Intersessional Consultation on New and Emerging Digital Technologies
Machine-readable formats: Plain text · JSON
Transcript
Transcripts available through this tool are created by using automatic speech recognition and are not official records nor official documents of the United Nations. Official records and official documents are available on the Official Document System of the United Nations. Learn more
If people could take their seats, I think we're going to want to get started quickly. I have the pleasure to announce that as a belated gift for this important intersessional, I'm told that we have both interpretation, which was unexpected, and webcasting. So I hope that's of benefit to people. And just to let you know, it's going to be available for today's session. Wonderful. Well, it's great to have such a nice full room of people who are engaged on these important issues today. My name is Peggy Hicks, and I'm the director of the Thematic and Special Procedures Division. And it's a pleasure, really, to welcome all of you to the intersessional meeting convened pursuant to Human Rights Council Resolution 59/11 on new and emerging digital technologies and human rights. At the beginning, I'd really like to thank all of you participants for joining us. I think it really reflects the truly global and multi-stakeholder nature of the conversations that are absolutely essential to ensure that technological innovation advances human dignity, equality, and human rights. I'd also like to express our deep appreciation to the members of the core group of this resolution. We have had such effective collaboration and discussions with all of you over the course of this, the work on these issues, and of course, in particular, to the Republic of Korea as the penholder of the resolution. We're very glad to have representatives of the core group with us today, and we know some of them will be speakers throughout the program. And of course, a warm welcome to you, Ambassador. for the Ambassador and Representative of the Republic of South Korea, Ambassador Yoon, who will give her opening remarks following me. So, why are we here? Why is this intersession all such a crucial one in so many ways? We know that digital technologies are increasingly shaping nearly every aspect of our lives. They influence how we access information, education, healthcare, employment, public services, and democratic participation. They create extraordinary opportunities for innovation and development, yet they can also generate or amplify risk to privacy, equality, and non-discrimination, freedom of expression, labor rights, civic space, access to remedy, and many, if not most, other human rights. Technology will continue to progress. And the challenge of our time is to ensure that that progression occurs in a manner that respects, protects, and fulfills human rights throughout the entire lifecycle of the technology. Human rights cannot, in this sense, be treated as an afterthought, nor as a downstream compliance exercise. They must be integrated from the earliest stages of design and development through deployment, operation, oversight, and eventually decommissioning. This principle has been, as many of you have heard me say those words before, I have to admit, because it's been a consistent thread through the work of the Human Rights Council and our Office on Digital Technologies over a number of years. But the Human Rights Council, through successive resolutions, has progressively developed a stronger understanding that human rights considerations must accompany technological innovation at every stage, and that both states and business enterprises have essential roles to play in making this a reality. Today, much of our discussion focuses on one of the most important tools that are available to help translate these principles into practice: human rights due diligence. Human rights due diligence provides a practical framework for identifying, assessing, preventing, mitigating, and addressing adverse human rights impacts before they become entrenched. States, companies, and regional bodies increasingly treat this as a core part of the corporate responsibility to respect human rights under the UN Guiding Principles on Business and Human Rights. And it's also an important means through which states can implement their own human rights obligations in the context of digital technologies and artificial intelligence. This evolution is significant. The reality is that states are regulators of technology, but they're also procurers, investors, developers, deployers, and users of digital technologies, like all of us. Governments themselves increasingly rely on AI-enabled systems and digital infrastructure to deliver essential public services. As a result, questions of accountability, transparency, oversight, and access to remedy arise across a growing range of state activities. At the same time, as we know, technology companies continue to exercise enormous influence over how digital technologies are designed, developed, and deployed. So we need to ensure that both companies and states integrate human rights considerations into their decisions to have effective governance. The state business nexus is particularly important. As public and private actors are deeply interconnected through procurement arrangements, digital infrastructure projects, financing mechanisms, and public and private partnerships. As such, human rights responsibilities cannot be viewed in isolation. Strong governance requires attention across the entire ecosystem. The first panel this afternoon will examine precisely these questions. It will explore how states and companies can operationalize human rights due diligence throughout the lifecycle of digital technologies, and how this framework can strengthen accountability, transparency, participation, and access to remedy. As discussions on digital governance increasingly turn toward implementation, the focus of the panel will be on how principles can be translated into practice. The second panel will move us to the topic of emerging challenges of agentic AI, recognizing that our discussions take place against a backdrop of rapidly evolving technological capabilities. Recent advances in artificial intelligence are accelerating the development of increasingly autonomous and agentic systems capable of initiating actions, coordinating tasks, and interacting with other systems with limited human intervention. High Commissioner Turk has warned that this is a watershed moment in which decisions made today on how advanced AI systems are designed, deployed, and governed will have consequences across multiple generations. Yet we are not yet meeting this enormous challenge. Governance is struggling to keep pace, and we have not yet put in place effective measures for AI systems already in use in fields such as the administration of justice. And there's much hand-wringing these days, but yet no clear path forward to global measures to reduce risks of frontier AI model development. Yet, as the High Commissioner outlined in his recent letter, there's consensus around some of the key steps that are needed, which include, first, ensuring independent, technically competent verification of agentic capabilities with access to models, documentation, and testing environments, and mandating ongoing human rights due diligence. Second, requiring reporting serious incidents in agentic AI testing and deployment, and empowering independent bodies to investigate such incidents immediately with full access. Third, requiring companies to have safety plans in place that meet agreed standards backed by liability regimes when the absence of such safeguards causes harm. Fourth, ensuring that responsibility for the actions of AI agents remains with identifiable humans, including by not allowing AI systems to hold legal personality. And finally, effective state responses need to require active and ongoing engagement by civil society to ground efforts in the experience and knowledge of those on the front lines of AI impacts. These, for us, are the essential preconditions for innovation that is sustainable, trustworthy, and beneficial to society. The second panel will then explore these existing governance approaches and how they can be applied and adapted in context of increasing autonomy, complexity, and scale. It will consider how and what accountability structures are needed, what meaningful human oversight means in practice, and what additional safeguards may be necessary as AI systems become increasingly capable and agentic. None of us can do this alone. Governments, companies, engineers, researchers, civil society, international organizations, and above all, the communities affected by these technologies, each see a different part of the picture. When rules are written without these voices, they tend to reproduce the inequalities, blind spots, and harms we hope to address. When people are brought into the process, the results last longer and carry more legitimacy. That's why we're here today. Today's meeting forms part of a wider effort across the Human Rights Council to address what digital technologies in AI mean for human rights. Recent council initiatives show how pressing these questions have become and point to a growing recognition that human rights must remain central to governance frameworks now taking place. As the High Commissioner reminded us this week at the Rights Act Summit, human rights must serve as our compass in navigating these uncharted waters of technological change. Managing the risks is the bare minimum. Ensuring that decisions about technology remain grounded in human dignity, public participation, and accountability is our fundamental task.
Thank you.
And with that, I'll turn to our fellow convener here, Ambassador, over to you.
Thank you. Excellencies, distinguished colleagues, let me begin by thanking OHCHR for organizing this meeting and all of you for being here today. It's a pleasure to join you at this multi-stakeholder inter-session meeting pursuant to Human Rights Council Resolution 59/11 on new and emerging digital technologies and human rights. Needless to say, the pace of technological change is accelerating, AI systems in particular. are being developed at an even greater scale and speed with a transformational influence on almost all aspects of our lives. Without doubt, these developments offer significant opportunities. At the same time, they raise important questions about the obligations and responsibilities of different actors involved in developing and using these technologies. As AI becomes more deeply integrated into public and private activities, how can we ensure that these obligations and responsibilities are fulfilled in practice? In this respect, international human rights law provides a normative foundation. International human rights law establishes states' obligations to respect, protect, and fulfill human rights, and these obligations apply in the digital environment as well. As the Human Rights Council has repeatedly affirmed, the same rights that apply offline also apply online. The UN Guiding Principles on Business and Human Rights also provide an established framework in this regard. They set out the state duty to protect human rights, the corporate responsibility to respect them, and access to remedy. Central to this framework is human rights due diligence. through which the corporate responsibility to respect human rights is put into practice. Building on this framework, the recent OHCHR analytical report on human rights due diligence for AI and state obligations, mandated by Resolution 5911, further clarifies how states can fulfill their human rights obligations through due diligence in the context of AI. It also identifies challenges relating to public procurement, state-business partnerships, and systematic and cumulative human rights impacts. Distinguished colleagues, as AI systems take on more complex tasks with greater autonomy, questions about how human rights obligations and responsibilities can be fulfilled in practice come into sharper focus. Unlike AI systems that primarily respond to human instructions, agentic AI can autonomously plan and carry out multiple steps to achieve a given goal with limited human intervention. This greater autonomy increases the risk of accountability gaps. It can become more difficult to trace how decisions and actions unfold. determine where responsibility lies among the actors involved, and maintain appropriate human oversight. So how can we ensure that human rights due diligence remains effective as AI systems become increasingly autonomous? This is a question that deserves closer examination through today's discussion. As UN High Commissioner for Human Rights, Noted at the Rights Act Summit yesterday, we still lack common rules to guide the development of AI. Given the speed of this development, we are at a critical moment for developing common principles and practical safeguards based on established human rights norms to guide the design, deployment, and use of AI and other emerging technologies. Today's meeting offers an excellent opportunity to identify areas of convergence and explore how human rights standards can be applied in addressing the increasingly complex challenges facing states and businesses, including those arising from developments such as agentic AI, drawing on the experiences and expertise of states, businesses, civil society, and other stakeholders. Looking ahead, the Republic of Korea intends, in close consultation with our core group partners and other member states, to bring forward the next resolution on new and emerging digital technologies and human rights at the Human Rights Council next June, taking into account the rapid evolution of AI and other emerging technologies. We hope that today’s exchange will make a substantive contribution to that process. I look forward to a rich and constructive discussion. Thank you.
Thank you very much, Ambassador. And those conclude our opening remarks, and then we will now take a quick technical break to switch over to the first panel. Thank you.
Thank you.
Good afternoon, excellencies, distinguished delegates, colleagues. So I have the pleasure to moderate the first session on human rights due diligence, looking at this as both a conceptual framework for ensuring that digital technologies are governed in accordance with human rights law, but also as a practical process that can operationalize human rights obligations and responsibilities. We often hear that, or we often discuss human rights due diligence in the context of corporate responsibility. It is, as we've heard, a central pillar of the UN guiding principles on business and human rights. But due diligence is also deeply rooted in international law, and it's also one of the principal ways in which states discharge their obligations under international human rights law to respect, protect, and fulfill human rights. And many human rights obligations at their core are indeed due diligence obligations. What does this mean? It means that states are not simply required to react when harm has occurred, but they're also required to take reasonable and effective measures to prevent foreseeable violations, to identify risks before they turn into harm, and to establish safeguards in law and through institutions that are capable of protecting rights in practice. So, OHCHR, as you heard already from Peggy, has consistently argued that both states and companies need to undertake human rights due diligence in relation to digital technologies. And I'd like to mention two reports that we have as an office on the practical application of the UN Guiding Principles and the activities of tech companies, including in the context of AI. You will find reference to this in the concept note. And we also have a report under the new and emerging tech resolution that was presented earlier this year on how human rights due diligence relates to state obligations under international human rights law. And there we look at some of the specific features of AI that set certain expectations for how human rights due diligence should be carried out and what that should look like in the context of AI. So I'm fortunate, we are fortunate today to be joined by a distinguished group of speakers who have spent many years thinking and engaging on this topic, bringing expertise from government, civil society, academia, and the private sector to help us understand the concept of human rights due diligence for digital technologies and for AI. We'll also touch on one pressing question in digital governance today, what we referred to often as the state business nexus, and what human rights due diligence requires when states increasingly depend on private actors to design, develop, and operate digital systems that shape the lives and rights of millions of people. With that, I have the pleasure to give the floor to Her Excellency, Ambassador Mendez Escobar from the Permanent Mission of Mexico for opening reflections.
Thank you very much, Madame Tagini. Is that right? Thank you. Thank you very much, and a very good afternoon, everybody, colleagues, friends, panelists, Inria, Lita, Linda. It's a great pleasure to be appearing on this panel. I'd like to make it clear to you that for Mexico, this area, human rights and digital development, is a priority area. We're working on national legislation in this sphere. We're very grateful to the OHCHR for organizing this dialogue on the application of due diligence in human rights to digital technologies and artificial intelligence. This debate is particularly timely because it invites us to move from principles to the practical implementation of our obligations and responsibilities. Allow me to share three thoughts from the Mexican perspective. First of all, due diligence in human rights should cover the entire life cycle of digital technologies. Due diligence cannot be understood simply as an act of hindsight. It should begin from the design and development of systems, it should continue through their rollout and use, and incorporate permanent oversight mechanisms for accountability and reparation as well. And this is particularly important in the case of AI and other increasingly complex and autonomous systems whose impact can develop over time and simultaneously impact multiple human rights. The second thing, states should also apply due diligence when they use, develop, or acquire digital technologies. Very often, we hear about corporate responsibility, but states play multiple roles within the digital ecosystem. We regulate, we fund, we develop, we buy, and we use technologies. And as a result, requirements around due diligence should also guide government decisions. And in that vein, Mexico believes that public tendering represents one of the most powerful tools to promote responsible innovation. When governments acquire digital systems or tools based on AI, they should incorporate requirements over risk assessment, transparency, human oversight, data protection and effective access to reparation. Government tenders can become a decisive trenchant incentive for human rights to be incorporated into the design itself of technologies, and not only be incorporated into post hoc regulation. Thirdly, technological innovation should progress hand in glove with inclusion accountability and the access to effective remedy. Proper due diligence requires participation by those potentially impacted, particularly those confronting the greatest risk of exclusion or discrimination. And it also requires that we have accessible mechanisms to question automated decisions. This is extremely important. present claims or complaints and obtain reparation when human rights are impacted. Innovation and accountability are not mutually exclusive. On the contrary, they are mutually reinforcing and they are vital to build public trust in new technologies. Colleagues, experts, Mexico has consistently promoted the business and human rights agenda and continues to strengthen its national frameworks in this sphere. Throughout 2026, we reactivated our work to draft a national policy on business and human rights, pursuant to a wide-ranging consultation process with government, private sector, unions, civil society, and international bodies. The National Human Rights Program for 2025 to 2030 reaffirms Mexico's commitment to a human-centered digital transformation which is compatible with human rights. The guiding principles on enterprises and human rights continue to provide the most robust framework to guide this task. The pillars contained therein to protect, respect, and repair continue to be fully relevant to respond to challenges raised by digital technologies and artificial intelligence. Allow me to conclude with a thought. The challenge we confront is not simply that of developing the most advanced technologies. Our shared challenge is to ensure that technological innovation strengthens human dignity, broadens opportunities, and contributes to the effective enjoyment of human rights. Due diligence in human rights is one of the most important tools to attain that goal. Mexico reiterates its commitment to continue contributing to an inclusive, responsible, and human-centered digital governance. Thank you very much, Ana.
Thank you very much, Ambassador. And thank you particularly for highlighting what we can think about as really the precautionary approach when it also comes to our institutions and the mandates needed for them to be able to provide the oversight over what is often cumulative impact of digital technologies. I now have the pleasure to turn to Ms. Ilya Siatitza from Privacy International, where she's program director and senior legal officer. Ilya, you have the floor.
Thank you, Azeen, and thank you to the conveners for bringing us together. I speak here on behalf of Privacy International, a civil society organization based in London, where we work globally with partners to challenge companies and corporate, companies and governments' abuses of data and technologies. The question put to us here today is how due diligence can move beyond procedural compliance to meaningfully prevent harm, ensure accountability, and provide effective remedy to affected individuals and communities. And my short answer to that is due diligence is meaningful only if it can change the outcome. It must be able to stop a deployment, rewrite the contract, and give a harmed person someone to hold to account. I want to make three points on how to get this there, building on the High Commissioner's analytical study on human rights due diligence and AI. First, due diligence has to follow the data. AI systems are built on data, run on data, and keep learning. from data. The study rightly notes that data provenance and governance shape AI outcomes, yet in our experience, these are often under-examined in risk assessments. Data protection impact assessments are essential, but not enough on their own. This is where due diligence is key. Given that AI often builds on existing data sets, due diligence must ask harder questions. Where does the data come from? And was it lawfully obtained? Should it be collected at all? What stops it being used for other purposes? And does collecting it put people at risk? Once data exists, it can be breached, sold, used to infer sensitive information, or repurposed. Data and systems also move between uses. The UN Secretary General has warned that the inherently repurposeable nature of AI poses challenges to oversight and accountability, and urged states to examine the blurred lines between civilian and military AI. The study is also clear in that regard that due diligence apply whether for civilian or military purposes. So safeguards cannot depend on whether a system is called civilian or military on the day it was assessed. Intended uses and the safeguards against incompatible ones must be reassessed throughout the life cycle. Second, that brings me to the state business nexus. States are no longer simply buying products. They enter long-term partnerships in which a company runs Quite often, part of a public function, and often gains access to data to develop its own products. The study, again, rightly calls for these partnerships to be fully subject to due diligence, including contractual safeguards. And Privacy International safeguards to public-private surveillance partnerships can help put their recommendation into practice. I have two short examples in that regard. First of all, assess before you sign. Impact assessment should be done and published before the contract is awarded, not ticked off after it. And the level of scrutiny should follow the risk, not how much the value of the contract was. And the second one, keep the right to walk away. Contracts need termination clauses, interoperability, and no lock-in. A state that discovers harm should not be locked in by exit penalties or by a company that holds its data and cannot take it away. And this brings me to my last point. Due diligence must be able to say no. That was already highlighted also by the ambassador just before me. The study, again, is clear. Where risk cannot be adequately mitigated, states should refrain from deployment or prohibit specific uses altogether. When due diligence reaches that conclusion, that is a red line, not a risk to be managed, and it must have consequences on procurement, non-contract, and a public finding. In some cases, the answer is already clear. Where the consequences are lethal and meaningful human judgment is removed, as with anti-personnel autonomous weapons, the line has been drawn. The UN Secretary General and the UN High Commissioners have called for those weapons to be prohibited. Closer to everyday life, there is facial recognition in schools. It means collecting children's biometric data in a context where consent cannot be freely given because of the power imbalances within schools. It chills children's development and less intrusive alternatives exist. Our analysis concludes that it should be banned, and that was also confirmed, recommended by the UN Special Rapporteur on the right to education. These two examples are closer than they look. Facial recognition first used at military checkpoints, has found its way into schools and hospitals. The blurred lines run both ways. In short, meaningful due diligence follows the data, is written into contracts, and can say no. As the next panel will turn into agentic AI, this becomes even more urgent. Opacity cannot justify weaker protection. It increases the duty of care. Thank you very much for your attention, and Privacy International is there to support the Office of the High Commissioner and states in this work. Thank you.
Thank you very much, Ilya, for giving us very concrete examples and recommendations, and also for reminding us that human rights due diligence in some situations means red lines. I now have the pleasure to turn to Ms. Anita Gurumurthy, executive director of IT4Change, advocating on digital justice, data governance, and platform power. Please, you have the floor.
Thank you so much. It's a pleasure and honor to be here. I'd like to start off where Ilya left. by pointing to the fact that AI, as we know it, is a production paradigm and a knowledge infrastructure. It's just not a product or a tool. So we have a range of rights that concern society that get implicated. Epistemic rights, especially with respect to the public domain, economic justice, labor rights, environmental rights, all of these. And therefore, human rights due diligence needs to step up to this imperative of going beyond an individualistic conception of harm. What due diligence means is that we are building the techno-institutional baselines for all of these things, for labor, for environment, for democratic accountability, for market fairness, for how innovation can be shaped and backed up with proper accountability, and we do really need to be on guard that due diligence does not become a cop-out, just compliance paperwork for normalizing bad technology. Due diligence cannot be a liability shield. What is important to understand is that there is, in the study referred to at the start of this session, some important benchmarks the recognition of the systemic and cumulative harms associated with digital technologies, the need to extend due diligence across technology lifecycle, and of course, the precautionary principle. But what I'd like to do is to look at what all of these three things that are so significant in the report mean in terms of the foundational premises. The first is that Downstream correction as a default is no longer an option. Of course, the use of AI does require post-market monitoring and correction, but that is not to be confused with the due diligence that goes with the duty of care of actors who build, deploy, own the AI. And all of this is about the organization of production. It is not about tinkering with individual harms. The second premise is that AI needs to be seen as a class of technologies, and in the case of AI, that is dual or multi-purpose, the thresholds for due diligence and liability have to be different. For example, due diligence and liability in nuclear technologies are structured to manage long horizon risks. strict regulatory oversight, and catastrophic third-party damage alongside specialized statutory liability regimes, because the harm can really be exponential. In the case of the EU law just one year ago, the Grok episode really threw up the need to move beyond simplistic risk thresholds to active baselines. It pushed EU lawmakers to tighten prohibitions on intimate deepfakes and treat generative AI harms as a systemic enforcement priority. The third premise is that institutional modalities for due diligence need to be set up. We need new institutions that set public standards for incidents reporting and for adjudication. For example, Ireland's Regulation of AI Bill 2026 establishes a register of independent adjudicators. Similarly, the EU AI Act relies on independent third-party notified bodies for pre-market conformity assessments. And the fourth premise is a very important premise for civil society. A societal right to participate in AI-related policymaking is necessary, and we do need to engage women and gender minorities as well as young people. We've seen that corrections in the market usually happen for commercial law, such as copyright, but the dignity of personhood is as important, if not more. What I'd like to conclude with is, how do we operationalize due diligence with clear rules? Rules need to look at a few things. How do we measure real world impact? Evaluation has to focus, as my predecessor said, on actual outcomes for the people affected by the tech, especially vulnerable groups. And these have to be based on constitutional values. Secondly, what are the hard prohibitions and risk tiers, definitions of what are unacceptable uses, and what are pre-deployment obligations of high-risk systems? Thirdly, how continuous post-deployment monitoring will actually happen? For instance, where are the verifiable inventories? What is incident logging all about? Is there red teaming when capabilities or data start drifting? What are the internal controls? And in fact, the scientific panel on AI has made a very, very good list of such evidentiary basis. Fourthly, How will a product or application be decommissioned if unforeseen damage occurs? And finally, as was already pointed out, where does the actual accountability lie? In one of the conferences I was part of, someone gave a beautiful analogy, and I would like to bring that analogy here. No one who owns a dog and takes a dog for a walk can claim, "Oh my God, my dog went mad and I could not stop it from biting you." There is a duty of care, and therefore, accountability is extremely important, particularly if it's state-owned infrastructure. In fact, there is a tragic abuse of public digital infrastructures in India. For example, Telegram, the app, the Telegram bots act as accessible storefronts for generating synthetic intimate imagery, non-consensual synthetic intimate imagery And it's often monetized through our public unified payments interface. So the payments take place through local payment intermediaries who are opaque, but the payment is actually being routed through the DPI. So can you imagine the banking system being abused for non-consensual synthetic imagery? So I think that liability rules, unfortunately, seem to be a closed chapter for now, but we really hope that if the court cases are anything to go by in the United States for addiction and mental health, the black box of opacity can be opened up. There is a huge power imbalance here, and we do really need some presumptions of the burden of harm to be taken up by the powerful. And finally, a note on public procurement. which is a primary lever where high-risk AI rules and deployer duties take concrete shape. We require proof of lawful training data, as was already mentioned, and we certainly need to center the data rights of people, especially indigenous people, especially the minorities. And we do need termination rights in mandatory model clauses. Also, there is a deep irony here. Increasingly, states are using AI to get into the procurement process. So we really need to guard against this. And when we use AI to screen the bids, you know, we have to be really careful. My last point is that we look at inclusion and accessibility, and we do need, therefore, to walk the talk as states and use open source tools, interoperable platforms, and licenses that will allow public scrutiny for public procurement. De jure rights are not enough. I would like to point out that in the international political economy, trade rules cannot act as barriers to look at the code in the public interest. The public needs the rights to look behind the black box and look within the black box. Thank you very much.
Thank you very much, Anita, for taking us quite a few steps into the operationalization and also for emphasizing the process and the the inevitability of making sure that we include also affected groups in decision-making and in assessing impact. I now have the pleasure to turn to Ms. Linda Bonio, founder and CEO of Lawyers Hub, who works at the intersection of law, technology, innovation, and public interest advocacy. Linda, please.
Thank you so much, Chair. I speak from my experience at the Lawyers Hub working across Africa on tech governance, and also from the first United Nations Global Dialogue on Artificial Intelligence Governance, which I co-chaired the human rights cluster here in Geneva in July. In most of the world, the state sees artificial intelligence as a buyer. In Africa now, we are seeing many of the countries trying to build this system on the basis of digital sovereignty. So most of these governments actually do not, they buy them for welfare, for identity, for tax, border control, and policing. Many of these laws work because there is peace, but we are seeing procurement in times of crisis. Whether it's immigration, whether it's war, or what we saw last week in the United States on grounding of airplanes. And so we have states that sometimes have perfected, you know, crisis procurement as, you know, a pullback on regulations around this. So for buying states, in this case, it's the ability to inspect that system, to control what it runs on it. and to leave this vendor without losing the public service. The African Union Data Policy Framework of 2022 and the African Union Continental Artificial Intelligence Strategy of 2024 set that direction for the African continent, and I was privileged to be part of the drafting of both of these documents. In the UN report on business and human rights this 2025, there was insights around capacity building, on the fact that many of these public procurers really lack the technical and human rights expertise to vet and test these claims. And so that confirms to us that the state may hold the duty of due diligence, but most of the knowledge rests with the companies. So in the work that we have done in Africa, we launched the Africa AI Governance Index this July at the Lawyers Hub here in Geneva. where we scored all 54 African states on 80 indicators. And we realized that no state reached the advanced band, and most of the states are very high on strategy, but really are outrun by delivery. And so there's a capacity gap that's there. And most of this is explained by compute. Africa has 18% of the world's people, but less than 1% of the global data center capacity. UNDP found that 5% of Africa's AI talent has the, does not have, only 5% of Africa's countries has the talent that it needs, you know, which is really indicating that many states do not have access to graphic processing, and many of this talent or systems are borrowed from the vendor. That means dependency and reliability on the vendor systems to actually test and do due diligence on these systems. I'd like to talk about three cases emanating from Kenya, where I come from, that are really useful to look at in this regard. Kenya's social health authority used a machine learning model to predict household income and to set health insurance premiums. An investigation by Lighthouse Reports, Africa Uncensored, and The Guardian rebuilt the model and found that it charges the poorest households too much and the wealthiest too little. So the authority released the model's inputs only after the ombudsman pressed them to release this data. And so one pregnant woman's appeal to the automated line was refused without reasons. And so we are seeing the impact of AI in public infrastructure. And we see this, we don't know this until you talk to the ombudsman and people who actually, we have not built their capacity for this specific age. The nexus also runs outwards. We've seen over 71 governments, including many of the countries in the EU, subcontracting visa applications to VFS. A second Lighthouse report also investigated with 14 media partners, and they saw that there were automated negligence and registrations and African applicants. really were being hidden behind a private player when they are trying to access public resources. So really, the state's duty does not transfer with the contract. We need states to actually be able to say that we take responsibility for these technologies. But then also finally, in 2021, the High Court in Kenya quashed the rollout of Vuduma Namba, which was the national digital identity card, because there was no data protection impact assessment that has been done. We wasted over six years of the time not implementing digital ID because of the failure of states to be actually, you know, to do due diligence before deployment of this particular technology. So the court determined that a human rights impact assessment for artificial intelligence needs these features. So I would like to mention that I think we must make due diligence a precondition for procurement. We come from a continent where actually procurement is for really long-term projects. When we have procurement in crisis and we have donations coming in, in terms of AI, really, we see subversion on the requirements for this particular due diligences. Secondly, we need cross-border remedies. A person excluded from a welfare system that, you know, comes from another country, cannot reach the company that designated. For instance, I was denied a visa to Germany, and they asked me to appeal in Germany. And whatever was written for my case was not ideally, even my passport, it was written to an Indian. These AI systems are making mistakes, but people do not have the platforms to actually show up in Germany or whatever country to be able to state their case. So we need some cross-border remedies that will be able to do this. And the High Commissioner already wrote on 14th September that the few states hosting frontier companies, they carry a particular obligation to hold them accountable. So what do you do with Silicon Valley companies that are actually harming people in Kenya, Ray-Ban glasses from Meta, that are dearly impacting people across the world? need cross-border remedies. And finally, we need to finance capacity building. You know, these due diligence standards really can happen to wealthy states, but what do we do with low-income states? The Global Digital Compact asked for financing options for capacity building. In September, the Secretary General asked the General Assembly to advance a global fund for AI capacity building. And on the Digital Cooperation Day, I was there, there were discussions around how do we fund capacity building for technologies so that we on the same page as we talk about due diligence, we train procurement officers, independent assessors, and to share regional computing capacity in this sense. I think these models exist. At the Lawyers Hub, we've been conducting the Africa AI Policy Summer Schools, and we've held editions in Ghana, in South Africa, in Egypt. And the next one, we're looking at AI and elections. What elections could technology ideally mean in terms of procurement? We've seen this with Kenya's 2007 and 2013 elections, and the US has faced the same sort of allegations around Venezuelan technology that was used in Kenya. How do we then transition and make sure that the world is on the same page? Because what is happening in Africa will happen to you, because tech companies have figured out that they will perfect the technologies of the rest of the world and then use it abroad. As the UN transitions to a new Secretary General, I believe that we must be able to look keenly on what these conversations could contribute to the high-level review or the Global Digital Compact at its 82nd session, and the co-chair summary of the Geneva dialogue records convergence that human rights must be protected throughout the life cycle of artificial intelligence. I ask that the summary report of this meeting really carry this. procurement due diligence, the cross-border remedy, and capacity to all these forums that we look forward to next year, including the global dialogue on AI. Thank you.
Thank you very much, Linda, also for giving examples of how to mitigate the capacity asymmetry between government and companies. But it was also refreshing to hear examples of how national institutions, ombudsmen, and courts can actually check and perform their duties with respect to how government uses or deploys digital technologies. So with that, I now have the pleasure to turn to our last speaker, Ms. Jen Hering, Senior Manager for Human Rights within Microsoft's Responsible Business Practices team. Jen, very happy to have you here, please.
Thank you. Excellencies, distinguished guests, thank you for this opportunity. Microsoft is committed to building trusted and trustworthy technology, upholding the highest standards of trust, safety, and responsible business practices as we develop and deploy technology. Microsoft has been invested in driving responsible approaches to business for many years, and was one of the early adopters of the UN Guiding Principles on Business and Human Rights, which were unanimously endorsed in June 2011, and we published our first global human rights statement that same year. As a company, we're committed to respecting human rights as we evolve our technology, operate our business, and engage our global supply chain. And this work is guided by the UN Guiding Principles. We follow its operational expectations for policy commitment, due diligence, stakeholder engagement, and remediation. In 2018, we conducted a or commissioned a human rights impact assessment of artificial intelligence, and that work served as a foundation for the establishment of our Office of Responsible AI, which developed our six responsible AI principles and our responsible AI standard that guides that implementation. We refreshed that human rights impact assessment in fiscal year 2025 with a focus on generative AI, and our responsible AI transparency report provides more detail on this work. As has been stated, states have a duty to protect human rights, and companies have a responsibility to respect human rights. We believe that responsible engagement can, in many cases, better advance human rights than disengagement. Where we operate in challenging environments, we take a principled approach consistent with our commitments to human rights and responsible AI development and deployment. and use our leverage and voice to help prevent and mitigate adverse impacts. We advocate for public policies and laws that promote innovation and align with our human rights commitments. We do so through regular multi-stakeholder, multinational, regional, and local engagements, and we also engage with experts to inform our efforts. In this AI moment, The human rights framework, as the High Commissioner so eloquently said yesterday, provides a guide for building and deploying trusted technology. Responsible government adoption can be a powerful catalyst for AI diffusion. When governments provide digital services, they help create demand, build trust, and accelerate broader adoption across the economy. Microsoft's AI diffusion research finds that leading countries are those that invest early in digital infrastructure, AI skilling, and government adoption. And our May report highlights examples such as South Korea, where government policies helped accelerate AI use across schools, workplaces, and public services. If diffusion is about getting AI into the hands of everyone, then human rights help create the conditions for diffusion. by building trust, legitimacy, and public confidence in how AI is deployed. Thank you.
Thank you very much, Jen, also for affirming that the centrality of the human rights framework for your company and for government as well. With that, you have now heard from all our panelists. We have about 15 minutes, I think. Or do we have more? Let me double check. No, we have more. We have 30 minutes before the next panel. So the floor is open for reflections, reactions, questions. Russian Federation, please.
Good afternoon, colleagues. Some comments in connection with this interesting discussion. Russia is convinced that the international community's approach to the protection of human rights in cyberspace should be based on the International Covenant for Civil and Political Rights. Article 19 of which guarantees every person's right freely to express their opinion, except in cases established by law that are necessary to respect the rights and reputation of others, and/or to protect state security, public order, health, or morality of the population. When protecting human rights from criminal encroachments, we have the United Nations Convention Against Cybercrime developed at Russia's initiative. This is the first universal legally binding instrument in the field of criminal justice in the digital environment, and we call on all countries promptly to sign and ratify it. Currently, international efforts are focused on expanding the scope of the document through the development of a protocol on additional offences covered by mechanisms of international cooperation provided for under the Convention. Special attention will be given to the protection of personal data. We urge all countries to support the constructive work on the protocol and contribute to its development. Distinguished speakers, as an example of at the national level, we want to point to the following. We attach great importance to improving legal regulation of the human rights and freedoms protection in the context of ICT development, including through artificial intelligence. And in particular, the relevant provisions are enshrined in the National AI Development Strategy for the period to 2030. as one of the binding principles for the development and use of AI. It also reflects the need to ensure the security and safety of this technology, which implies, amongst other things, protecting users from unlawful use of AI technologies. Thank you.
Thank you. Brazil, please.
Thank you so much, Madam Moderator. It was a real pleasure to listen to all the panelists today, and I think it's a very timely discussion to be holding it this Friday with the coming forum on business and human rights next week. I understand we're going to have a number of discussions on this topic too, and with the discussions on autonomous weapons happening right now as well. So just to begin, it's also interesting to ask how BTech and OHR has been engaging with the process happening next door. And just a number of interesting things that were said. I picked up especially on what Anita mentioned. She began by saying how many human rights implications we're talking about here, really. We're talking about privacy, employment, development. gender and race bias, and even environment. And since it's such a far-reaching human rights consequences that we're talking about, so one reflection also that I would like to put is how far-reaching across the value chain of digital technologies can we actually require due diligence to go, right? And how do we prevent companies to skew their responsibilities by third parties. I'm wondering here if due diligence has to go all the way, for example, to the data centers and where does it stop? And these kind of questions, and particularly about market concentration, which is also an issue that's been here pervasive when Madam Linda asked in how do we prevent the companies in Silicon Valley It's a small number of countries that control the models, a small number of companies, the computing capacity and the data that all the rest of the world relies upon. So yes, due diligence frameworks, we believe, should address market concentration, access to information for public authorities, very important. Ms. Anitha also mentioned about transparency and lifting the black box. So, yes, this is why Brazil believes that the UN Guiding Principles and the negotiations towards a legally binding instrument too that's been happening here are complementary. And so I suppose, apart from all these considerations and reflections, I would like to pose a question. What minimum human rights requirements should states include in the AI procurement contracts? so that developing countries can negotiate on a more equal footing with large technology providers. Thank you again for your very valuable insights.
Thank you very much for the reflections and for the questions. We'll take a few more, and then I will turn to the panel for their also potential responses to those questions. Arminia, please.
The moderator, and I would like to echo my colleague from Brazil in thanking you and the OHCHR for convening this very timely discussion. Indeed, the timing couldn't have been more suitable. And as a co-sponsor of the Human Rights Council Resolution 59/11, we highly value the Council's continued engagement on new and emerging digital technologies and human rights. We thank the Republic of Korea and the core group for their sustained leadership and commitment to fostering inclusive dialogue and international cooperation on this very important agenda. Our engagement builds on our country's long-standing efforts to advance gender equality and meaningful participation of women and girls, including in digital sphere. I think we can all agree that while technological transformation offers unprecedented opportunities, Challenges like algorithmic bias, discriminatory outcomes, and unequal digital access illustrate that the benefits and risks of digital technologies are not distributed equally, with significant implications for women and girls. It is against this background that Armenia, together with the cross-regional core group, has introduced Resolution 63/13, which was adopted at the 63rd session of the Council. and which seeks to advance gender-responsive, inclusive, and human rights-based approach to digital transformation. Recognizing the importance of reliable data, scientific evidence, and a deeper understanding of gender-differentiated impacts, it seeks to foster dialogue between technical and scientific expertise and human rights perspectives. To this end, it envisages a panel at the 68th session of the Human Rights Council. and invites the independent scientific panel on AI to contribute to this meeting, and we look forward to this very important contribution and the voice of science. Madam Moderator, we see considerable potential for these initiatives to complement and reinforce one another, and we look forward to working closely with all interested delegations, partners, building on our shared objectives and respective contributions to advance coherent human rights-based approach to digital technologies, including artificial intelligence. Once again, thank you very much for convening this very important discussion.
Thank you very much. I see Poland, please.
Thank you and good afternoon, colleagues. Thank you very much for convening this important meeting today. Poland has been the supporter of this resolution, and we are glad to see so many important interventions done during this panel. Today, we support a human-centric approach in which AI and new emerging technologies protect fundamental rights and freedoms while enabling innovation, public sector transformation, and sustainable economic growth. This priority has been present also in other resolutions where Poland is the penholder of, namely the good governance resolution presented at the last session of Human Rights Council. where AI is part of the discussion on its deployment in governmental administration. Among other things, Poland also, together with other member states, is engaged in the introduction of human rights-based approach in the standard development process in ITUT, the standardization sector of this organization. And in that sense, Also considering that ITU is where various stakeholders meet regularly, we would be interested to hear the views of the panel how this initiative could be strengthened and how we can work together to make this happen. Thank you.
Thank you very much. Colombia, please.
Thank you very much, moderator, and my thanks go to all of the panelists for their presentations. Just a short question: after the resolution which is the basis for these deliberations, what do the panelists think the human rights mechanisms should offer to the diverse conversations taking place right now? Particularly, what do they think the next step for the HRC would be to address the problem before us? Thank you.
Thank you very much. If there are no further questions at this stage, Please allow me to turn to the panelists for any reflections on the comments and questions that were raised. We had one on how far reaching across the value chain human rights due diligence applies, and also a question on how to deal with market concentration, and also minimum requirements for procurement contracts. We had questions on technical standard setting. and the role of human rights mechanisms, and I believe potentially other questions that I may not have noted correctly, but I'll let you choose which ones you would like to answer, and we can start with Anita and go to my right.
Thank you. It's quite overwhelming that all of us feel that this issue is so significant and we want to move with a certain spirit of urgency, and therefore I think conversations and dialogues will help us move the pieces. The first thing I want to say is that for very, very long, and this is in response to the extremely vital points that Brazil made, the conversation for very long has focused on red lines. I think the conversation should shift to baselines. And what I mean by this is that existing frameworks establish important requirements for identifying, assisting, preventing, addressing adverse human rights impacts. The baseline framework proposes minimum non-negotiable conditions governing AI production and deployment, which encompasses labor rights, ecological sustainability, democratic accountability, and equitable organization of AI markets. Without these four baselines, there is no point in thinking about red lines, which just simply identify unacceptable practices or production arrangements. There is a difference between the two. We have to start with the outcomes, measurable outcomes in terms of human rights, achievements, attainments, and the intermediary rights towards those attainments will comprise the baselines. And considering that unfortunately, we don't have an international consensus in the form of hard law around these specific baselines, I think it's extremely important for countries to look at their constitutions, their constitutional principles, to arrive at these baselines. For instance, one example that we often think is vital is subcontracted work. Fundamental labor protections are necessary in the AI value chain, not only to cover lead firms and their contracts, but also the subcontracted arrangements that lead firms engage in. to ensure that working conditions as well as remedies for workers and subcontracted arrangements in the value chain exist. That is a baseline. Similarly, holistically, we have to account for environmental impacts of AI infrastructures, including data centers, and provide for the prevention and mitigation of impacts before projects can proceed. I can actually give you a set of examples like this and would be very happy to share it about what baselines can be. The second thing I want to say is, what do we actually consider a checklist when we look at public procurement? Unfortunately, formal procurement processes do not treat harm specific to digital technologies, algorithmic opacity, unauthorized secondary uses of data, and vendor lock-in. I think vendor lock-in has been ignored for a very, very long time, and actually, infrastructural dependencies of developing countries really, I think, is rather a critical emergency. And state collaborations, therefore, with the private sector are not even conducted under formal procurement channels. So there are several commercial confidentiality, scrutability, looking at training decisions and algorithms, code bases, looking at conditions for vendor lock-ins, And there are some wonderful guidelines. UNESCO, for instance, along with a couple of organizations, including mine, has evolved a set of guidelines for public procurement. And therefore, I think that these may be worthwhile looking into.
Thank you very much, Anita. Ambassador, please.
Muchas gracias.
Thank you very much, dear moderator. I have three thoughts on the basis of the questions that have been raised. The first to my colleague from Colombia. This is the golden question: what are we going to do after the Human Rights Council resolution? To be honest, the issue of new technologies and artificial intelligence is more than a cross-cutting issue now in the international agenda. The Human Rights Council is dealing with it, the WHO, WIPO, the ILO. And the issue here, really, the challenge, is how can all of these processes around conversations and debates and recommendation drafting, how can they converge? How can they avoid overlap and duplication, and how can they be mutually reinforcing? This is an enormous, an enormous challenge on an issue which is completely cross-cutting. And so I think we need to focus a lot on how the states contribute to this debate and conversation, because obviously it's not up to the sectariats of the organizations to get the ball rolling. We need to ensure that we leverage synergy in this area. The second thing is what the ILO is doing. This year, I think it was in May or June, during the ILC, the International Labour Conference, we adopted Convention 193 on decent work and digital platforms. I would invite you to read the text of that if you can, and even watch the debates. Because the kernel of the conversation was whether or not the algorithms have human intervention, human oversight. This is the key issue in that debate, in that convention, and it was more or less resolved. But we now need to see implement, well, first of all, we need ratification. And we also need secondary regulation in the member states to uphold it. But there was an enormous amount of work done at the ILO on that, and I really do invite you to look at that convention. And then we were talking about human rights, obviously, and I would note that this panel comprises 100% women talking about business and human rights. and artificial intelligence. And I think that sends a good signal, it's a good sign we're making progress. And I would say, and I didn't mention this in my initial intervention, Mexico is very committed, obviously, to this issue also from a perspective of migrants, communities, indigenous peoples, LGBTIQ+ community, persons with disabilities, because it's those groups in a situation of particular vulnerability who, at the end of the day, this debate, which is still running, will impact in the most negative fashion. And I think we all need to have that on our radar, that it's not just the ramifications on the regular citizen. It's the ramifications for groups in a particular situation of vulnerability. Thank you.
I turn to Linda, please.
Thank you very much. Thank you for the questions. I would like to build up on the ILO Convention 193. I think it offers a really good basis on the work that's happening around the world, especially with AI workers. I think we need to believe them, we need to empower them. And we need to look at how to institutionalize multi-stakeholderism, especially in this new era, because the talent and the work is resting outside of the state. And so as we need to institutionalize them as much so that we move faster. And the secondly, I think, is on conditional finance, to your question on how do we move post the resolution? We learned this in Kenya from the Huduma Namba incident, the digital ID. We didn't have a data protection framework in the country, but the resistance brought and gave birth to the privacy law, establishing a data protection commissioner. And at that point, people took the opportunity to again define artificial intelligence and autonomous systems in 2019 and say, even without an AI framework, we will not have autonomous systems. that do not have human review. There must be a human in the loop. And I think that is what we need in terms of pushing forward these conversations. The World Bank then was put to task and told, you will not fund digital ID until there are frameworks for us to work with, especially privacy frameworks. And so I think that the next step will be conditional financing, that without these frameworks, we are not financing certain projects. Third, there's weakening of institutions, public institutions. We are seeing this everywhere. We must strengthen the institutions that work on these issues, from the judiciary to the offices of ombudsman, to the data protection officers and human rights commissions. That is very necessary. And finally, we must fund mechanisms for human rights. Human rights organizations are struggling across the world, especially with funding cuts. How do we innovatively continue to fund human rights? Because the world needs it. We cannot let human rights be run by big tech companies that are working out of philanthropy. We must look at how governments can ideally be part, human rights work must be part of government funding this particular work, but not having them on different sides of the table where human rights organizations are enemies of government. I think human rights organizations are friends of society, they're friends of the future. That is how we make sure that due diligence is done. So could we think about a particular threshold of annual budgets or global GDP that could ideally go into funding this kind of assessment? I think that will bring stability. It will also make sure that human rights organizations are not behold by big tech companies that require certain, you know, concerns to be met as a prerequisite to funding them. That will reduce the dependency. Thank you.
Thank you, Linda. Over to you, Ilya.
Thank you. Lots of great questions and a lot of Issues that have already been addressed, and it's a great, really honor and pleasure to be engaging in this discussion with all of you because we see how we are all bringing together and we are facing the same challenges. The first one I wanted to address was one that was already mentioned already by the interventions and before, is the multiplicity of the conversations that they're happening at the moment. We have been participating at the local conversations, we are here today, and we are also next door at the review conference discussion, where we are hoping your colleagues will soon come back with a mandate for a legally binding instrument to regulate lethal autonomous weapon systems. And one concern that I have there is multiplying the same conversations, the questions about autonomy, about oversight, about ensuring that the humans remain in control are happening across all these rooms. And how do we ensure that actually the same core fundamental principles are implemented across all those implementations? Because otherwise, the risk is that the one is going to undermine the other, and we're duplicating work as well, and especially when it comes to AI. This is essential because the foundational models, the infrastructures, the data, and the companies are the same. Some implementations change, and there are some specifications there, but the fundamentals are all the same. So this is where I think the human rights framework and the human rights, the Office of the High Commissioner more specifically as well, have a core role to play to make those bridges. and to ensure that all these conversations are not undermining human rights protections at their core. The second point, which draws from that one as well, is remedy and transparency. It was already mentioned a couple of times how in order to ensure that those responsible are held accountable, there needs to be transparency with regard to what systems have been used, which companies are involved, and which And which authorities are the ones that need to be consulted? And don't get me wrong, when I'm saying that, I don't mean that what it is often lately the case where then the various stakeholders involved will be allowed to point the finger to each other with regard to who is responsible. Because at the end of the day, the decision to deploy an AI system for offering a public service continues to lie with the state. So the example from Kenya that was given before about the system actually giving unfair rates. to the different for social benefits, it is still on the state to ensure that and go back and remedy this situation and compensates those impacted fairly. And I think this is something, especially as a conversation around and I quite often tend to anthropomorphize the systems. We need to remember that, that we need this, we need to hold those that they're deploying to account and they remain accountable. And the last thing I wanted to say, there was a question, a very important question on the value chain and how far should we go? It goes back to my initial point. There are certain aspects when it comes to concerns with data centers. These are, there are some core aspects that they are specific to those parts of the life cycle. So we can break It doesn't, we can break and make the assessment specific to the different steps we are talking about and we are looking into. And that can also help alleviate some of the workload. Like not every local council that adopts a chatbot to help citizens in their Dialogue with citizens should be going and going back to do a due diligence impact assessment that goes back to the data centers. This is something that needs to be centralized. They don't have the resources to do that themselves. So these are things, a few of the thoughts I had from listening to also my companions and all of you. I'll stop here, and I'm looking forward to continuing the discussion. Thank you.
Thank you very much, Ilya. Before I turn to Jen, we had one point also on monopoly from Linda. So Linda, please, that point, and then we turn to Jen.
Thank you very much. I was going to maybe comment on the monopoly question. I believe it's time for us to support startups in action and in policy. There is no way that you will work against monopolies by breaking them down only. We have seen the experience from the FTC in the US with Lina Khan, Lina Khan and the work that she was doing. Ideally, it works for a period of time, but if the same capital, the same support does not go to startups, we risk building big companies that turn out to be, you know, a problem, especially in terms of national security that we have seen. There are models that we've seen across, for instance, in terms of local ownership for big companies. I think the model in India, Kenya had 30% in saying that for you to begin a company, you must have a local. coming in, that really triggers the ecosystem. But without that, we see gentrification coming from these big tech companies. I also believe that we must stop criminalizing entrepreneurship, especially in certain countries where entrepreneurs are seen as a problem and they want to make money. I think there's need to make sure that as many people as possible are in entrepreneurship, that balances the system. We also must take an ecosystem approach to due diligence. We cannot just work with companies only, we must work with startups, we must work with local organizations where people actually understand what's happening, then we have a great approach. The Silicon Valley approach towards having Stanford working, VC companies working together, you have companies coming together. If we don't take that ecosystem approach, then we risk having these really great monopolies that we have no idea of how to change it. And finally, we must celebrate initiatives and startups that are building and people that are doing due diligence. In celebration, I think, in the end, and in scoring sometimes, that celebration then gets us to a point where we have tiny companies that are ideally doing the right thing. And in Gen. Z language, they say, we made the right people famous, and they buy from them.
Thank you, Linda. And finally, Jen, please.
Thank you for the questions. Per the question regarding value chain, aligned with the expectations in the UN Guiding Principles on Business and Human Rights, Microsoft does apply the UNGPs across our entire value chain as we engage our supply chain, as we operate our own business, and as we develop and deploy technology. And the UNGPs apply whether our customer is a government or an enterprise or a consumer. And you can find more information about how we do that in practice in our Human Rights Transparency Report. Thank you.
Thank you, Jen. If the next panel could indulge us with one more additional minute as this, as we don't want to break the flow of the comments here, I'd like to turn back to Anita for an additional point.
Thank you for the indulgence. I won't take long. I wanted to make two points with respect to Columbia's important provocation. I don't have the answer, but if you did ask me about my wish list, I'd really like to say that the whole debate on human in the loop, I think, is really not tinkering with the problem on a case-by-case basis. It's not about leaving it to individual coders or programmers or these algorithmic minds in a company to resolve issues one by one, but it is about a systemic intervention at two levels. One is how are we shaping the architectural rules, so when you build, you actually have to build by certain rules. And this includes techno-social standards, techno-legal standards. So The whole idea of bringing the human in is about architectural rules, it's about systems, it's not about the human being alone. The second is institutionalized accountability systems, which means these are not fuzzy, these are actually concrete. So this is actually what we mean when we say human beings should be there in the mix. It means society should be there in the mix. The second point I also want to make is that unless we have a societal approach, we will end up through huge cases in the court where, through torts law, we are awarding billions of dollars to some parent who may have lost their child because of a mental health issue. We do not want companies to settle outside the court with these aggrieved populations. We really want a rule of law framework which goes beyond commercial law. We want a societal law.
Thank you very much, Anita. And with that, I thank all the panelists for their interventions and for sharing their insights and experience with us, and also for the interventions from member states. Thank you very much. Now, panel two will take over, moderated by my colleague, Isabel Ebert.
Good afternoon, everyone. Welcome to the second panel. My name is Isabel Ebert. I co-lead the BTEC project with my colleague Natalie Stadelmann. This panel moves from the setting of the scene in terms of the core frameworks that we discussed that we have available through international human rights law and the UN Guiding Principles on Business Human Rights. as a conceptual basis, and now we are moving to the, and what now? I think it's on many of our mind today. We are facing pressing problems in terms of agentic AI governance. We've heard which human rights due diligence requirements for states and companies are necessary to ensure human rights are protected against adverse impact linked or stemming from AI development and deployment, and what this means in the state business nexus with regard to digital technologies. We are now turning our attention to the "so what now" piece of the conversation. How can we move from principles to practice, from insight to impact? The recent revelations about problematic behavior of AI agents bring current challenges in oversight of agentic AI to the forefront. They have highlighted also the urgency to find effective pathways for solutions to ensure agentic AI is designed and deployed with human rights safeguards in place. Therefore, the focus of this panel debate will rest on human control, accountability and responsibility, independent oversight and verification, and how due diligence applies where systems can act with increasing autonomy and at scale. We are discussing on the basis of the following guiding questions: What human rights risks and governance challenges do increasingly autonomous and agentic AI systems pose? How should human rights due diligence evolve to address systems that act with greater autonomy and scale? What constitutes meaningful human oversight and control in the context of agentic AI? And how can accountability be maintained across increasingly complex AI value chains and ecosystems? What can technical standard settings contribute to the governance of increasingly autonomous AI systems, particularly in advancing safety, accountability, and respect for human rights? And what should the role of standard setting bodies be in that regard, also with including regulators, auditors, researchers, civil society, and national human rights institutions, and each playing their respective part in oversight and accountability? And last but not least, under which circumstances may additional safeguards, restrictions, or governance measures be necessary to address risks that cannot be sufficiently mitigated? We have a stellar panel today, to say the least. I have to say, in particular, for a rainy Friday afternoon. To start with, to my left, Ambassador Jaya Rantham. The floor is yours, please.
Thank you very much. I am very pleased to join this distinguished panel here today to share Singapore's experience and be part of this conversation. AI is posing a unique challenge. The technology cycle is qualitatively different from anything we have experienced in the past, in three ways: it is broader, it is faster, and it is more holistic. We have all heard dystopian concerns about its perils. We also know its promise in making real differences to our lives. The key question is what do you all believe? For us in Singapore, we focus on first principles. This means at the heart of the matter, AI must serve humanity, and never the other way around. The fundamental challenge is: how do we manage these risks without stifling innovation? As AI becomes more capable on our behalf, we must correspondingly introduce more robust safeguards. We must be clear about what AI is allowed to do, ensure humans remain in control and accountable where it matters, and continue to strengthen our ability to understand emerging risks and respond where problems occur. This will help build trust in AI as a technology that serves public good. I would like to share Singapore’s approach. This is what we believe to be the central question: How can we realise the benefits of technology and connectivity while managing the risks that come with them? What we look to do is to ensure that innovation is matched with assurance. and that our policies remain flexible and adaptable to new developments. There is no one-size-fits-all approach to making AI safer. We must calibrate limits according to risk. We must consider the sensitivity of data and systems involved, the degree of autonomy given to AI system, and the severity and remediality of any potential harm. Rigid requirements on AI developers could prove counterproductive. Companies will limit their cooperation or information sharing. Regulatory frameworks should be done in consultation with industry, with our common interests in mind. Otherwise, they will simply not be implemented. We believe that delivering on AI safety in a way that protects innovation is a shared responsibility at all levels. First, governments must set the right baselines, develop standards, build capabilities. that organisations cannot efficiently build on their own. Second, on their part, the tech companies must build security and assurance into the systems they are developing from the beginning. Third, researchers can help us identify capabilities and risk. Fourth, operators, when deploying systems in real-world environments, must ensure that systems remain safe in real-site conditions. Finally, all must work together to find out what works, work together to ensure success can be scaled. In Singapore, we have already been implementing a holistic and cooperative approach on the ground. Let me give you some examples. First, Singapore has a model AI governance framework for agentic AI and a starter kit for testing LLM-based applications to provide guidance on managing risk and maintaining human accountability. Second, our Cybersecurity Agency released its Securing Agentic AI Addendum earlier this year, which was developed with industry, government, and international partners. The addendum provides practical guidance to help organisations secure increasingly autonomous AI systems. Third, our Global AI Assurance Sandbox facilitates real-world testing of risks of generative and and agentic AI, including the robustness of their safeguards. For example, we have worked with Google on our AI Agent Sandbox, which helps us better understand how agents operate in real-world settings and how to use these insights to inform how they are developed, deployed, and governed. We recognise that different countries have different approaches to AI governance. We should therefore be practical and flexible, rather than seek to establish rigid frameworks that which do not get widespread adoption. We have a common interest in making AI safer while continuing to harness its benefits. Singapore is keen to contribute to the international conversation in this regard. We have convened the International Scientific Exchange for AI Safety, where AI experts from academia, industry, and civic organisations have developed the Singapore Consensus for AI Safety research priorities. We partner international bodies like ISO and IEC to standardise methodology for testing generative AI systems. We are active participants in the UN processes. We are especially proud to have a Singaporean scientist in the Independent International Scientific Panel on AI, and we are pleased to have participated in the inaugural Global Dialogue on AI Governance here in Geneva. Our Digital Minister is also on the AI for Good Global Commission. And of course, we were very happy to join Norway and Finland’s call for control of frontier AI models in September. Dear friends, throughout history, new technologies have generated both excitement and anxiety. People have welcomed their potential to improve lives while fearing the disruption they may bring. Singapore remains committed to doing our part so that the benefits of AI can be enjoyed safely. and also by many, not a few. Thank you.
Thank you, Ambassador. I'm turning next to Ambassador Kment from Austria. You have also joined the Finland and Norway call, as I have heard. The floor is yours.
Thank you very much, Chair. Good afternoon to all. My respect for your perseverance on a Friday afternoon is really much appreciated. And I'd like to say that it's a pleasure, of course, to contribute to this roundtable, and I would like to thank everybody who's been involved in the organization of it. We've seen a dramatic acceleration in national and international discussions about AI safety and security, and also about the geopolitical dimensions of AI. looks like that a global consensus is emerging on at least one point, that is the urgency for frameworks and guardrails for the further development of advanced AI, ensuring that the deployment of artificial intelligence systems does not affect the core of what defines our humanity. That's also very much in line and the key message of Pobleus Magnifica Humanitas, which says, I quote, "In the era of artificial intelligence, when human dignity is threatened by new forms of dehumanization, ours is the pressing duty to remain profoundly human." End of quote. That's the, I think, very pertinent appeal from the Pope. The human rights risks posed by increasingly autonomous AI systems are very well outlined in the OHCHR report on human rights, due diligence, and state obligations. They provide an excellent basis for our further work on this issue. AI systems, again, a core message of this document, fundamentally reshape how violations and harms manifest. Reliance on automated decision-making in public administration, in security, and in the military can entrench errors and biases that are difficult to correct over time. The report also outlines that this is a crucial point in my regard how the capabilities of today's advanced AI systems already pose very serious concerns related to human dignity. challenging thus the foundational assumption of human rights law and international humanitarian law. And it is with this urgency in mind that we have joined the call for the Frontier AI Models, which today reunites 32 countries and the European Commission. We believe that in light of increasingly autonomous and agentic AI systems, Global common standards for AI governance that strengthen transparency, including the reporting of safety incidents, and include human rights by design are necessary. And we also support that UN member states explore creating an international AI institution able to set standards, enable verification, and facilitate coordination among states when thresholds are crossed. Companies should develop transparent safety protocols, including mandatory pre-deployment testing and independent evaluation and verification. The most urgent need for international cooperation is a shared understanding of what we are managing. Our cooperation should focus on interoperability, on shared terminology, assessment methods, standards of evaluation, and verification, as well as mutual recognition of results where this is possible. Diverging approaches create legal uncertainty for the industry and gaps in oversight and accountability. The UN cannon should play an important role. in this regard, to streamline already existing initiatives, achieve conceptual clarity, and enable closer cooperation on safety and ensure compatibility with international law. A common language in all these efforts are and should be human rights and international humanitarian law. Most of the problems we are starting to see with advanced AI systems, both in civilian and military context, can be mitigated by ensuring human rights due diligence throughout the AI lifecycle. And this starts with the collection of data, includes investments in development, and ends with deployment and decommissioning. And a comprehensive approach here is very, very important. We must make sure to engage in some of the less traditional fora, including on technical and security-related spaces, to carry human rights into these discussions. Ignoring the potential human rights consequences risks steering us into a dystopian future, whereas including human rights safeguards will make sure that all of humankind profits from these technologies. I'm glad in this regard that OHCHR increasingly engages on discussions on security-related aspects of AI, including emerging technologies in the military domain. And just across the hall, we are engaged in the very difficult negotiations on autonomous weapon systems. This is a process that's been going on for 13 years. where Austria and many other countries have tried to bring in human rights aspects, and it's been very difficult because we still have a silo approach on these issues. I think it's very important that we break this up. And we've been arguing for some time, also more broadly in the Human Rights Council work, that a clear distinction between general purpose and military AI is no longer possible. They are already fused on several levels. When it comes to frontier models, they are adopted for defense purposes, that are adopted for defense purposes or to infrastructure such as clouds or data centers. Even more important is the use of data. Military AI applications run on civilian data, the collection of which is, or rather should be, governed by human rights. Vice versa, we have seen data collected in conflict with no guardrails whatsoever being used to train commercial AI models. The reality is that most AI technologies, whether initially designed for civilian or military use, are inherently dual-use and easily repurposed, and that blurs the line between civilian use and military applications. This raises concerns about privacy, equality, non-discrimination, liberty, security, freedom of expression, peaceful assembly, due process, and the right to an effective remedy for discriminatory outcomes, all of which are standards under human rights law. And still under international human rights law, lethal autonomous weapon systems, as I have just pointed out before, very clearly would violate the right to life, undermine human dignity, and fail the requirement for legal accountability. So I will end on this, and I look forward to the discussion. Thank you very much, Chair.
Thank you, Ambassador, in particular also for emphasizing the urgency of human rights safeguards by pointing to the very salient risks for human rights in the military and AI domain. We're now turning to our civil society segment of today's panel. What would you add to the list of challenges in oversight of agentic AI and some pathways for solutions? I would like to give the floor next to Meredith Wade from the Business and Human Rights Center.
Thanks, Isabelle, and thank you to the great interventions of the panel thus far. We're off to a very good start. My name is Meredith. I work with the Business and Human Rights Center, as Isabelle mentioned. And we systematically track corporate facilitated human rights harms allegations in our database. And over the past two years alone, we have recorded more than 140 allegations of human rights harm linked to AI companies and their value chains. So a couple of these examples have already been raised within previous interventions. Upstream, we have logged cases where data center expansion is linked to the displacement of Dalit communities in India, and lithium extraction for AI hardware is threatening indigenous land and water rights in Chile. We have cases of data labelers in Kenya, Colombia, and the Philippines, where workers are describing constant exposure to traumatic content, unfair wages, little worker protections. And in the downstream, we have examples of AI tools generating sexualized images of women and children, and has already been mentioned, being deployed where possible crimes against humanity and war crimes are having AI facilitating those incidents and violating the right to life. So when we talk about frontier AI on this panel today, I want to make sure that we're also framing this within the context of the value chain, which the representative from Brazil very rightfully pointed out earlier. The issues that I just mentioned and all of these allegations of harm, many of them remain unresolved and unremediated, yet the industry is steamrolling forward with frontier AI. And now, to add to the list of challenges, as is the question, frontier AI companies are normalizing serious human rights harms in their S-1 IPO filing forms, as we've seen thus far with the case of SpaceX, which owns xAI and Anthropic, presenting these risks to investors as routine business risks. When a company can disclose a risk to humanity and still list on a public stock exchange, catastrophic risk is being absorbed into ordinary vocabulary of corporate finance. But when we look at this through the lens of the UN Guiding Principles, the fact that these companies are admitting to the extreme risk means that they need to actually prevent it and not just disclose it to investors. But that also needs to be added to the long list of other types of harms that many civil society orgs have been documenting and raising to companies now for many years. Another question that was posed was about keeping accountability across frontier AI value chains. And I think we need to start from the point where we recognize that something that's technologically novel does not make it legally exceptional. There are instances where courts are catching up. States attorneys general are suing frontier AI developments over their safety claims. We have wrongful death suits that are naming CEOs personally in the cases. We have positive reflections thus far on where we're going to go with fundamental rights impact assessments under the EU AI Act, but we can't wait for courts until after the harm is already done. The burden of proof needs to sit with the companies, the frontier AI developers. And they need to be able to prove to regulators, to their investors, and to the general public with verifiable evidence that they comply with the existing laws that we have on the books, including consumer protection, data protection, non-discrimination, labor and supply chain due diligence, environmental and intellectual property law, international human rights law, and humanitarian law, as has been raised. So I have a few calls to the governments that are in the audience. First, we need to make sure that legislators and regulators, not frontier AI companies, are deciding what's acceptable. Today, when we look at the behavior of frontier AI companies across the board. Incident reporting is very narrow. It's slow. It's hidden from public view. Human rights impact assessment disclosures are non-existent or very incomplete. We need legislators to further clarify these existing obligations on frontier AI companies, so again, so that not all the work is being done by the victims who are seeking remedy in courts. And we need regulators. that have real access, not just to the models, but also the competence to assess whether or not these frontier AI companies are effectively carrying out human rights due diligence across the entire value chain. We also need to be able to look at the financial incentives and the rules that govern how money is flowing into these companies, and how these frontier AI companies are able to maintain control. Many frontier AI companies have what are called dual class or multi-class shares, which means that once they do go public, it doesn't really contribute to meaningful increased transparency or accountability. The tech sector is not the only group that is involved in these types of share structures, but they're making this routine in what we consider to be highly risky areas where they're largely unregulated and the rules are under enforced for products that are being tested upon billions of people. And we have already learned our lesson for when we have very powerful tech companies testing out their products on the general public. It wasn't long ago that the International Court of Justice found that Myanmar had committed genocide against the Rohingya. And in 2018, UN investigators had said that social media played a determining role in that crisis. And investors, while the crisis was going on, were trying to raise for there to be impact assessments on disinformation and how the platform was governing harmful content. But those resolutions that were raised were not successful, also largely due to the dual class share and voting structure. So as we see now, again, with recent news, Anthropic is trading as of yesterday. And the co-founders retain more than 50% of voting power through one single share, which they say is meant to protect the company's vision. But also it's very clearly labeled in the investor documents that it, quote, "generally will not allow others to affect outcomes of how the company is operating." And this is extremely dangerous from our point of view for these frontier AI companies to maintain that level of control. And lastly, I would urge for governments to strengthen protections for whistleblowers and human rights defenders and make sure that stakeholder engagement is a key requirement for how these systems are designed, developed, deployed, and how continuous oversight and governance is maintained. We need to reframe the way that we think about how these companies are able to operate and deploy their systems. In a lot of states, protections for whistleblowers, which are key points of information in an environment which is otherwise extremely opaque. They are not receiving full protection for their concerns around all types of human rights related concerns. And there have been examples where there are allegations that companies have contractually prevented staff and former staff from criticizing the company for life. And states need to make sure that those types of clauses for employees are completely voided. So, in the interest of time, I'll wrap up by saying there are many things that frontier AI companies can do while states are getting the machinery going in order to adapt their ways of interpreting existing laws and adapting new laws and frameworks for regulation that clarify that frontier AI companies and tech companies in general are not the exception, they also have to abide by all of the laws that other companies have to abide by, that they also must adopt to the UN guiding principles on business and human rights. And where reasonably foreseeable human rights harms cannot be effectively mitigated, especially those that are being raised by the executives themselves and the engineers that are working in these companies, these tools should not be deployed.
Thank you, Mary. We're halfway through this panel, so I'm taking a little bit stock what we've heard so far. We heard about regulation, legislation, states needing to get active to define acceptable boundaries of agentic AI use. We've heard about applied solutions such as sandboxes, LLM testing, and red teaming. Then, of course, very much in the spirit of International Geneva, the need for international cooperation to stress the urgency also for common standards that we are working towards on the basis of human rights. Standards also in terms of applied evaluation standards, assessment methods, and ensuring interoperability. And then what the last speaker just added is, of course, also deleverage through the financial markets, right? So investor voices, investor pressure, looking at capital flows and also leverage points across the value chain. So turning to Alejandro Mayoral-Baños, the co-executive director of Access Now, what would you add to that already quite colorful list?
Thank you. Thank you so much, Isabel, and thank you for my fellow panelists. This is a very timely conversation, and especially with everything that is going on at the same time on the other side of this venue. So we at AccessNow are starting from a simple premise. Technology is never neutral, and AI agents are not neutral. Greater autonomy for AI must never mean less accountability for those who wield and deploy this technology. The risks before us are real. An agent acting across connected systems can affect essential services, private communications, critical infrastructure, and even democratic processes. That makes oversight urgent, but it does not mean we must start from zero. I want to challenge two assumptions. That agentic AI creates a human rights vacuum, and that an AI agent is a mysterious black box beyond scrutiny. We already have laws on privacy, personal data, and communication security. Where they apply, these obligations remain binding on the organizations operating AI agents. Autonomy is not an exception from the law. And an agent has components we can identify, its models and weights, software, data sources, permissions, connected tools, and hosting infrastructure. Open source software, open weights, and data transparency are different things. Not every component is fully visible, but we can demand access for independent verification and apply existing protections to the activities being performed. This leads to practical questions. Who authorizes the agent's actions? Who can revoke its access? Who keeps the records? Who is responsible when it causes harm? Consider AI features embedded in operating systems that capture screen activity. They can create searchable records containing private messages, even when those messages were protected by end-to-end encryption. In transit, the encryption is not broken, but privacy can still be undermined once messages reach a device without the other participant's consent. Now imagine agents able to read across applications, access files, and take actions. For journalists and human rights defenders, this is not abstract, nor are the wider burdens on shared networks and resources, especially in places with fragile connectivity. Then, I see three mutually reinforcing requirements. And this is an invitation for all the national states and stakeholders present today. First, enforceable rights-based regulation and continuous human rights due diligence. Assess risk through the entire life cycle, included before deployment, and whenever an agent gains new tools or permissions. Require testing of interactions across systems, Clear records of consequential actions, independent audits, and the ability to suspend deployment when harms cannot be controlled. A human approval button is not meaningful oversight if no one can understand or stop what happens next. Second, genuine multi-stakeholder governance. At-risk communities, civil society, regulators, national human rights institutions, and independent technical experts must shape these due diligence standards and its oversight. Technical standards can make safeguards verifiable, but cannot substitute for legal obligations. The civil society and the people affected should be and participate actively in building and shaping these due diligence mechanisms, not simply being invited to comment afterwards. Third, companies must build human rights into design from the outset, consistent with the UN guiding principles on business and human rights, guaranteeing freedom of expression and access to plural information. Limit permissions and data collection, preserve encryption, and establish clear responsibilities across developers, integrators, and deployers. When harms occur, people need effective remedies, not a claim that the AI acted on its own. Because such power is concentrated in few companies, obligations must primarily target them. We must avoid compliance models on the largest can afford. Oversight should be proportionate to risk, capabilities, and uses without entrenching monopolies. Long-term high risks deserve attention, but they cannot become an excuse to delay enforcement of protections we already have. The question is not simply whether there is a human in the loop, because that human may also be one deliberately causing harm. The question is whether the people, companies, and public authorities behind these systems remain accountable under human rights law. Thank you.
Thank you, Alejandro, also for the strong reminder that we're not operating in a human rights vacuum and there are standards that we can build upon. So, Alex Walden, the previous speakers have given you a lot of homework and inputs to deal with. We'd be very keen to hear how it looks inside one of the companies at the forefront of AI development and how you are currently managing human rights risks and where you see solutions going forward.
Do it? OK. So thanks again for inviting us to participate in this panel. I lead Google's Human Rights Program, our Global Human Rights Program. Google has been committed to human rights since our founding. In our first letter to stakeholders, our founders highlighted the goal to develop services that significantly improve the lives of as many people as possible. The company's mission from its inception to today has been to organize the world's information and make it universally accessible and useful. And building on that mission, we believe that generative AI and new and emerging technologies have the ability to benefit everyone by advancing the aim of boosting productivity, elevating creativity and expression, and the quality of people's lives. As models grow more sophisticated, people and businesses around the globe are transitioning from exploration to integration and finding new ways to put these tools to work in their daily lives. The transformational potential of AI is coming more clearly into focus, and from foundational advances in scientific discovery to clinical milestones in healthcare, the rise of agentic systems are capable of dramatically boosting anyone's productivity. So that's sort of our wide, our broader philosophy. But I also wanna talk more specifically about our belief that when it comes to AI, that we must be bold and responsible. For us, that means fostering innovation while prioritizing safety, security, and human rights. We're guided by our AI principles that were first published in 2018 and then were updated again in 2025. They emphasize that we need to be bold in our innovation, responsible in our development, and collaborative in progress. The human rights team, which I lead, works to advance our company-wide strategy. across human rights, across all of our products, including AI. And our sort of what we do and the ways that we incorporate that pragmatically are that we advise product teams on potential human rights impacts. We oversee human rights due diligence, and we engage with external experts and stakeholders. All of this work is grounded in our company's commitment to the UN Guiding Principles on Business and Human Rights, as well as things like the Global Network Initiative principles. As Google continues to bring AI to more products and services to help fuel creativity and productivity, we've established a robust internal governance structure that includes launch requirements, leadership reviews, post-launch requirements, and all of this is designed to support continuous improvement across our products and our structure for governing it. We continue to iterate on that internal governance structure. We carry out comprehensive safety and capability evaluations of our models, and that's complemented by external evaluations throughout the product lifecycle. Our external stakeholders include experts from academia, civil society, industry, and government. That approach helps us identify and mitigate risks before we deploy our models and on an ongoing basis afterwards. Human rights due diligence has to be building on that work and integrated into that work, the sort of safety and testing evaluation. So I want to provide a little bit more context for how we approach safety and testing evaluations. First, It's important for us to differentiate how we combine evaluations and red teaming. Our safety tests include two distinct and complementary practices. So there are structured model evaluations, which measure an AI model's capabilities and behavior in a structured and scalable way. Then we undertake red teaming. That's actively trying to break or bypass safety measures to find hidden flaws, vulnerabilities, or undesirable behaviors in an AI model or system. We also take precautions and have a robust approach to pre-launch safety testing for our frontier models. For certain models in particular, we conduct external evaluations for severe or frontier safety risks as described in our frontier safety framework. The frontier safety framework, which we've published, and I think that aspect of transparency is really important because it also helps provide more context to the rest of industry and to our stakeholders about how we think this work should be done. The Frontier Safety Framework is a set of protocols that ensure our most advanced AI models remain reliable through tested and aligned alignment with human values. The goal there is to understand the maximum capabilities and risks across five domains. Those domains are cybersecurity, sort of chemistry and biology, issues related to malicious actors, radiological and nuclear issues, again, related to high-risk nuclear radiological topics, sabotage, and harmful manipulation. So those are the five sort of core, most severe areas that those are meant to test against. We also engage with a variety of external organizations to get testing feedback. Two that I'd like to highlight are the government AI safety institutes for the US and the UK. We also have special external evaluators that we work with across a number of areas as well. From our perspective, it's important to ensure that we have a collaborative approach to security research, and we're engaging with experts to provide feedback around our AI products. And an important aspect of that is partnering across the safety ecosystem. So one or two things I'd like to highlight there are the Frontier Model Forum, which is a place where the Frontier Labs come together to address these issues around most severe risks and frameworks that we can deploy in order to address them, as well as some work we do with the partnership on AI, which is more of a multi-stakeholder forum for dealing with this. The last thing, there's sort of a lot to say on this. What I want to underscore is that all of the work that companies that we are putting into AI safety and testing should relate back relates back to human rights for us, and that we think that the rest of industry should be clearly sort of at least mapping the way that they think about safety to human rights. This is very much in line with what others on the panel have said. And we think that there is space for that both in the normative standard setting as well as in the standard setting that's happening at standard setting bodies. like the ISO. So I will stop there. Thank you for inviting us to participate here and look forward to more conversation.
Thank you, Alex, for giving us a peek behind the scenes. And also, i think some of the points we heard at the beginning of the panel were emphasized by you in terms of model evaluations, red teaming. We now have the pleasure also to have a technical, fairly technical expert, I would say, among us. Daniel Dobos, who joins us in his capacity as a co-chair for the AI for Good Impact Initiative. He can definitely say a few words about how the technical standard-setting community is thinking about human control, accountability, and also oversight of agentic AI, and where he would see a few pathways for solutions. Daniel Dobos.
Thanks a lot, Isabelle, and thanks a lot for the invitation, and also a big thanks for all my co-panelists for joining me. And allow me from my perspective as a scientist, have worked most of my career very close here at CERN as a particle physicist using AI, as an UN intern, and I will come back to that, but also as somebody who now works since several years in AI standardization, to give hopefully a little bit also refreshing view on what we heard so far. So, first of all, thanks, Ambassador Ratnam. It was really a great pleasure. And I think it was really important what you mentioned about the balance between risk and also innovation and how to balance that. And I had the pleasure to visit Singapore and meet some of the local experts in Singapore, Leslie Teo, which I think you mentioned being on the UN AI panel, Lawrence Teo. Many colleagues from Singapore AI, and this is a quite similar setting that we have also in Switzerland, having our own large language model, having experts. I think that really showed the need of balance and risk. Let me turn to you, Ambassador Alexander Kamenetz. I think the aspects that you mentioned of acceleration in defining what is the core of humanity and human dignity are very, very important. And I think this is also shared with some scientific colleagues, Johanna Prikker, also of the UN AI panel from Austria, I think is a very good example. And this really shows basically that this is a challenge, this acceleration that we see at the moment is one of the key challenges that we have. We have seen other technologies before. which might have been equally challenging if they would have been so fast, but this acceleration aspect of this technology is really one of the key challenges that we face all together. Meredith, that, I think it was really very, very refreshing to hear examples where we already see evidence. And I will come back to evidence, where we basically really see examples of evidence, and this is also, I think, what connects all the statements and examples you made to the International Independent Scientific AI Panel, which we already now mentioned several times. And I think it was really great to see that this kind of evidence gap means that we have to ask fast. And I think I really like the statement of everybody has to accept the law. Alejandro, I think it was really great to hear from you this kind of question of, you know, what can agents do by themselves and what is an AI agent when it acts on its own, and what then it really means. And last but not least, Alexandra Walden, I think what you mentioned, what stuck with me is this exploration, this transition from exploration to implementation. And you mentioned a lot of experts and oversight. Let me come back to the UN internship. I was a proud UN intern here once, and I would like to ask now a question. Do we really understand what agentic AI is? I don't. I don't fully know yet what real agentic AI is. And I'm not alone. Andrew Ng, one of the leading experts and teachers about AI, Even asked that question this year in Davos in the AI House, said, what does it really mean? And maybe he should know because he really helped to coin and distribute the term agentic AI. And still, he hasn't have a full definition yet and constantly changes his definition. The definition that he uses right now, I think, is a very good one, which says, There is a spectrum of agentic behavior. If you have something that where you tell a very clear work path, where you say, do first this, then do this, then do this, then it's more likely not very autonomous, it's not very agentic. If you give a very broad task, then it's highly agentic, highly autonomous. So let me use a metaphor that was mentioned in a previous panel about a dog, and I want to extend it. to a UN intern, which I was once also one proud of that. So now imagine I tell you that I have a new category of UN interns for you. Extremely intelligent. They have read everything, the entire library. They have read it. They work 24/7. They use a lot of energy, but they never sleep. They have the ability to reproduce themselves, and they can decide if they want to work alone or if they want to clone themselves, maybe 100 times, maybe 1,000 times, maybe 10,000 times. That sounds really appealing, right? That sounds like a very, very nice opportunity. But then, if you have that, would you then give this agent that maybe grew up on an island that has never lived among other humans, never really got socialized, because this is a UN intern that grew up on an island, very isolated, has never really experienced what it means to be a human, what it means to live in a society, what it means to make failures, to be accountable for it, would you start to give it access to your emails, to your bank account, to all your documents? to all your notices, maybe you wouldn't. For me, this is the best metaphor that I can come up with to understand what agentic really is, where the risks are. And we heard a lot of this risk, and I don't want to repeat it. All what I heard was really, really deeply thought of. But I would like to tell you a little bit more, and this I think is important, that we have to close the evidence gap of that. soon, because otherwise this acceleration will really challenge us. And I think the UN AI Independent Scientific Panel puts us really to the point with one sentence about this evidence gap. Sorry for that. So they summarize it very nicely saying that policymakers seeking to shape this, and they mean AI, this AI governance, are facing this evidence dilemma, but they need evidence to make informed consequential governance decisions, but by the time evidence exists, it might be too late to make them, as the evidence lags behind the pace of AI development. One aspect that we haven't talked about today, and this is why I would also encourage you to read to independent report, that preliminary report that have, and also the thematic brief, and this one on agentic behavior, is about AI children's rights. Because in all the discussions that I've seen, there seems to be some of the most consensus globally, also in AI standardization, as you asked, Isabel, that we have now scientific evidence, and this is where the UN Independent Scientific Panel on AI is extremely strong of, to collect this international evidence where we see that some consequence of AI, if used wrongly, then it can have negative effects. Also, the panel says very clearly that AI can very positively impact children's rights to information access, education, and expression, but under the appropriate conditions. We have to learn these appropriate conditions altogether. This is where we have seen now evidence if it's used in the wrong way, for example, in children's education, it harms critical thinking, it harms creativity, And this is, I think, one clear aspect of human rights where we all have a common interest to work together. And this, for me, could be a good starting point to learn from it. We have the scientific evidence that we need for it. We have a common interest to really protect the next generation. And we'll learn from the process how to work together on this. And then maybe we can also go to the more difficult, more controversial topics about AI. So this is where I would really emphasis to use the evidence that we already have, continue to identify the evidence gaps that we have, and then really support independent experts working on that. And this can be in all kind of fashion. This can be in big corporations. This has to be in independent universities. This has to be in collaboration internationally, and then also to start act on it. Thank you.
Thank you, Daniel. Always good to end on a strong metaphor that you can take home to think about over the weekend, how much liberty would give that hypothetical group of agentic grown UN interns. And also with a strong shout out to science and research and evidence-based policymaking, of course. Since we are quite good on time, we still have time also for comments and questions from the floor. I would like to open it here with. I know some of you might have had a long week, but please encourage everyone to find some energy to join the discussion. Okay, first European Union and then France.
As I'm intervening for the first time, let me thank you, OHCHR, for convening this topical intersessional meetings, and of course, warm thanks to all the panelists for their insightful presentations on subject of utmost relevance in the face of a fast-paced and far-reaching digital transformation. Let me say for the benefit of all that the EU supported from the beginning the resolutions on new and emerging technologies and human rights. that brought us together today, given that they put human rights at the center of our considerations, also in order to ensure that digital technologies benefit all. And in 2024 edition already, the resolution highlighted the need to assess the whole life cycle of digital technologies and to have regard to this. I think you know that the EU is both deeply committed to securing and advancing human rights in the digital age and a strong supporter to efforts to ensure sustainable and human rights respecting business conduct. But let me just say that we have notably championed a human rights based and human centric approach in the digital sphere and also adopted the approach of responsible and trustworthy AI innovation. In this regard, the EU also welcomed the adoption of the resolution at the last HRC on the implications of artificial intelligence for the promotion and protection of human rights. It was very important and timely that the Council raised its voice on the need to ensure that meaningful human rights-based AI governance keeps pace with its accelerating development and unprecedented risks to human rights. We, therein, or in our comments on the resolution, welcome the references to the need for human rights due diligence and meaningful human oversight and human control that I would like to re-emphasize here. And of course, also for multi-stakeholder participation in developing governance frameworks and the principle of transparency, explainability, and accountability. This call with due diligence chimes well with all the work that you have been referencing and that you've been doing at OHCHR, and also on due diligence, as the first panel discussed. And also let me reference that the Advisory Committee recommended enforceable regulation of the technology industry in reports earlier this summer. As recognized in the UN reports and studies, the EU has actually undertaken relevant regulatory efforts in the digital field, and it was often discussed and referenced to already here and in the last days. But for those who are not so familiar with this area of work, the 2022 Digital Services Act stands out as it established a regulatory framework to hold platforms accountable for harmful conduct, and the 2024 AI Act established rules for AI system. And then, of course, let me also mention the wider 2024 Corporate Sustainable Due Diligence Directive, which was adopted this year to the sustainability omnibus package. Speaking more specifically on the AI Act, I think it's really worth pointing out that it was the first ever comprehensive legal framework on AI, and that the aim is really developing, laying down rules to foster trustworthy AI in Europe, but then, of course, also by extension worldwide. The risk-based approach for AI developers and deployers is, I think, quite interesting, and it came together with a wider package of policy measures to support the development of trustworthy AI. And I think a little bit in connection to what was said also in the first panel, it is very important that in this approach, high-risk systems are actually, or there are nine types of high-risk systems which are not accepted to be entered in the market, and that there are specific stringent rules for other very risky systems that apply, and that perhaps helps you, us together in defining also valid and useful approaches going forward. Of course, the application of this rule, and I'm closing soon, takes place in a highly dynamic environment and is regularly reviewed together with stakeholders. And just this week, the Apply AI Summit gathered over 900 stakeholders to discuss a lot of pertinent subjects and showcasing AI innovation. And one of the subjects was, amongst other, providing new horizons on strategic topics such as frontier AI that we've discussed. discussing here as well. Before closing, allow me just to say two more, to make two more points. I think it is of the essence going forward that the work that is done in the multiple forums discussing these issues is better connected. And of course, we have a great interest that digital forums are grounded in full respect for human rights, just as the WSIS20 outcome document reaffirmed in December 2025. And this sort of mirrors what the Human Rights Council has done with digital technologies. I think in this respect, the joint statement on human rights and the work of ITU recently delivered in the ITU Plenipotentiary Conference was a very important step in this direction. And then, of course, in the same vein, let us recall this very welcome request in the AI resolution I just mentioned, that OHCHR holds consultations in the sideline of the second global dialogue on AI governance next year. We will be remaining very committed to supporting these efforts. Before closing, let me address one issue that wasn't really so much featuring in these discussions, but that we've discussed previously. And here I quote the previous EU Special Representative for Human Rights, who earlier this year addressed the argument of some that protection of human rights are costs, that they slow down growth and innovation, and that actually they add only regulatory burdens. And she referenced material also from international development banks that studies actually show that when people are safe or when institutions are more predictable, capital is willing to commit more for the long term. And this actually not hampers, but enhances growth opportunities and also improves the quality and the distribution of wealth. So I think that's worth pointing out that we're not in a binary situation, but that we need to go hand in hand in these efforts. So I'm sure that all of this is true for innovation and human rights, and we should develop them in parallel. So thank you so much. Thank you for the floor.
Thank you, European Union. Next on the list is France and Egypt. And please, France, you have the floor.
I wish to thank the OHCHR for organizing this intersessional meeting and the participants in this afternoon's panels for their very interesting and enriching presentations. France is active in placing digital issues and AI at the heart of international debates. States, academia, and business must work hand in hand to address the risks of AI. Innovation must continue to serve human beings and our collective security. France's President last week in Paris brought together representatives of states in the United Nations and business, researchers, and civil society to address this matter. The Paris Call for Security in Artificial Intelligence reiterates our shared interests and collective responsibility in AI, making it safe, secure, useful, and accessible to all. The exchanges shed light on the consensus on, firstly, the need for a clear, comprehensible, predictable regulatory framework; two, the fundamental importance of transparency and risk management, as well as signaling incidents; and three, the absolute need for rigorous evaluation standards. Together with Spain and Kenya, we have announced a coalition to protect children in the era of AI, ensuring it is safe, reliable, and respectful for the rights of the child in line with the CRC. Thank you.
Thank you, France. Egypt, please.
Thank you, Madam Moderator. As I take the floor for the first time, I just want to thank the core group of the Human Rights Resolution 59/11 and the Office of the High Commissioner for convening this meeting, and we thank the panelists for their informative presentations.
Egypt attaches a great importance to this topic.
And just to be brief on building on the discussions we heard today, I'd like to put three questions to the panel. The first one, where should we draw the line on autonomy when an AI system can decide either someone receives a public benefit or advances in a recruitment process? Which decisions should require human approval before they take effect, and which evidence should lead a government to withdraw?
Or suspend a deployment.
Second, who can act when something goes wrong? A public authority may be reactive for that decision, while the supplier holds the records needed to explain it. What should suppliers be required to provide so that authorities, particularly in developing countries, can in an independent way test the system under local conditions, investigate the complaint and resolve the harmful decision? The final question is: What should follow from a public review of technology companies? If a review identifies a serious risk to human rights in one country, How can it lead to a corrective action wherever the same risk arises? And how can the countries where these systems are used or the people affected by them help set the review's priority and submit evidence directly? Thank you a lot.
Thank you. Please, if you could also just identify yourself when you speak. Thank you.
My name is Anita. I was on the previous panel, and I am from IT for Change, an organization in India. I want to make two points. Firstly, to extend the argument made by the EU, and in fact state that compliance with human rights is not part of any inconvenient costs. In fact, an important step forward should perhaps be not to have a binary approach at all between innovation and inclusion or innovation and human rights. I think for too long, the idea of balancing the two has been out there in the discourse, and it is not a good way to go forward. I think we already know enough, and the evidence of the harms of a particular paradigm of big AI is all around us. So I would, in fact, urge that we go beyond the idea of balancing innovation and human rights. We talk about human rights in relation to innovation because we have done that all along, even with respect to other domains, including, let's say, pharmaceuticals and drugs. So I would like to submit that digital technologies are not really an exception. The second is I would like to also caution against framing AI harms around good use and bad use. That is too simplistic because harms happen without bad intent in AI. AI interacts with society, and systems that are built for helpful goals, like improving efficiency or hiring, can still cause discrimination or errors when trained or flawed or biased historical data sets are involved. Similarly, a single good use can still cause harm because the deployment may benefit one group, but may create negative side effects elsewhere, and therefore harms do happen even with a single good use. Similarly, harms evolve over time and across contexts. So, you know, it's a post-market monitoring approach that's really important. We need to stay agile because tools that work safely in a setting can completely fail if the data sets, for instance, not meant for the developing world, you know, are tools are being implemented in cases where the data comes from elsewhere, and this can create cascading effects. involving multiple people and systems. We have seen it enough, and I would submit that we have enough and more evidence on this. So I would like to urge that we do not really think about good use and bad use, but we find more useful ways in the law to categorize harms. For instance, who or what is affected? What is the intent behind building this? What are the mechanisms? How do we separate harms that have already occurred from probable risks and speculative scenarios? What is the system complexity involved? And there is more than enough literature on this. Thank you very much.
Thank you very much, Anita. Now, the gentleman from Brazil, please.
Thank you, Isabel. I'll try to be quick. It's Friday, and we are all I'm tired, but just building on a bit of what Egypt has said and also Anita just now, a short but definitely not simple question. So in the context of rapidly scaling agentic AI, what does the panel think that are the concrete criteria that should define meaningful human control in practice? Because since AI seems to be feeding more AI nowadays, so where does it stops and what can actually be qualified as human control from now onwards? Additionally, how can our international human rights mechanisms ensure that these independent auditing and technical standards remain genuinely inclusive of voices and regulators from the Global South? And since I have the opportunity, I would love to ask also to Ms. Walden to expand a bit on the recent update of Google of its AI principles in 2025, if she can expand a little bit more. She mentioned the 2018 ethical guidelines, and now regarding the pledge of not using AI for weapons and surveillance, if she can expand a little bit more on that. Thank you.
Okay, thank you. We will close the panel with remarks of 30 second maximum per speaker. We start with Meredith.
Okay, hefty task. I have many thoughts, so please feel free to find me afterwards for those who have asked questions if I don't answer them in time. Echoing what the European Union said, and Anita mentioned, and a number of others, we need to reclaim the word innovation. Innovation cannot simply be measured by what companies are capable of building and how quickly they can build it, or how much profit they can extract and hoard. It needs to be measured by whether or not technological advances are respecting human rights, protecting the environment, upholding rule of law, and really serving the interests of people and the planet, which is the interest of why we're all here today. And with regards to where we draw the line, I would say that we need to draw the line at where we reasonably are not integrating. adequate assessments of what the human rights risks are, and where we feel that systematically, we are not able to effectively have control or understand what the systems are doing and the ways in which they're impacting people. I would also encourage for other companies in the room, there are lots of mentions from corporations that are developing these systems about the importance of human rights and their human rights policies and the existing AI safety standards that they have. But upon quick review, none of the frontier AI safety standards mentioned human rights, actually. So we need for companies to take a lot more time and effort to do internal coordination and make sure that we don't have siloed policies and programs and agendas even going on within the same company. We need overlapping and communicating expertise from within in order to also help push for drawing clear red lines that we cannot deploy these systems where we cannot control them.
Thank you, Meredith. Alejandro.
A challenge, 30 seconds in a Friday. Okay. Very quickly, going back to Brazil question, I think it goes back a little bit of what I was saying in my remarks, is who authorizes those agents to start the work? And there are many, many steps in the cycle. the different permissions, who is keeping the records, and who is responsible of the infrastructure, and where all the data is being stored. So I think there is a lot of things of already placing responsibilities and accountability on AI agents. And I think going back to Anita's point, I think I really want to highlight your last point about breaking that good and bad uses of AI, I think that is not going to help us to move the conversation forward. And very connected to that is this idea that we cannot define AI agents. I think we can define AI agents and we can establish very clear human rights diligence with the tools that we have right now. And I think this long-term approach, or this catastrophic way of looking to AI, is putting a lot of governments to think about the future and not thinking about the urgent matters. And there is a lot of things that are happening right now, and we need to act on those right now. And we have already a lot of tools in order to decrease those harms.
Thank you. Alexandra.
Sure. I will also be here afterwards to chat about anything that I can't say in 30 seconds. But I think so much of what people have said resonates with the way that I think about that Google thinks about how human rights should be governing the way we think about AI. In particular, the reason our core sort of value around this that our leadership sort of trickles down across the company is bold and responsible is because that we need to innovate, and at the exact same time, we have to be responsible so that those things go together. It's not a balance, but it is a both/and. So I think that's an important point that has come up several times. On our AI principles, you know, we were among the first companies to roll out AI principles, and from the time that we rolled them out to 2025, significant changes happened across our company, across the technical capabilities, and what we learned about AI governance. And so our AI principles were really updated to make sure that we had a pragmatic approach to thinking about what we were doing and how we were governing AI in a way that was clear and understandable across the company. So it wasn't a move away from any principles. What it was, was a move to become more sort of pragmatic and implementable in the way that we were sort of doing trust and safety work across our AI infrastructure inside the company. And then maybe just the last thing I'll say is, and I'm happy to talk more about it afterward, is like, we believe that if it's illegal without AI, it should be illegal with AI. And so once you start there, that means that informs how you think about what additional regulation is needed, that it should have a sectoral approach, that sort of responsibility should be distributed based on the type of technology and the type of actor, whether it is a developer, deployer, or an end user. And so lots of these things can be incorporated in ways that are interoperable in global standards as well, so that companies that are working across jurisdictions can make sure they're living up to that responsibility everywhere.
Thanks, Alex. Ambassador.
Thank you, and thank you all for this really great and very enriching discussion. At the end of the day, our core values remain immutable, whether we are dealing with AI, whether we're dealing with any technology. How we defend it, how we protect it, that is the real discussion. I think my colleague from Egypt raised a very key point about responsibility and accountability. These are open questions that we have to deal and handle on a day-to-day basis. It has to be dealt with evidence as to exactly what we are looking at, exactly what we can do. And my final point is that, for me personally, the doctor's mantra of "do no harm" should be the starting point. Thank you.
The doctor's mantra is also the UNGP mantra of do no harm. Daniel, please.
Thanks a lot. Let me try in 30 seconds to address one question of exactly from our colleague from Egypt, where to draw the red lines. And this is an extreme challenge because the technology is changing so fast that it's very difficult to draw these lines. And some lines that we have been drawing today might be questioned again tomorrow because the technology has developed so fast that we have to rethink these lines. And I think this is why accountability, transparency is extremely important, so that we can understand that the red lines that we have been drawing yesterday are still valid today, or we have to rethink them and make sure that red lines are not threatening human rights of tomorrow.
Thank you, Daniel. And the last word, Mr. Ambassador.
Thank you. The distinguished The delegate of Egypt has asked the easy questions for Friday evening. Where to draw the line? Again, linking it to the discussions in the other one line. The issue of human judgment, human control has been discussed a lot in that context. It's very difficult. It's very unclear because international humanitarian law is geared to humans. So how do you apply it to these weapons? I would say we need to go back to, in order to address the human rights and IHL challenges, we probably need to go back to the fundamental questions of ethics about human dignity to try to find the compass there. what is still acceptable from a perspective, for example, the principles of humanity and public conscience. what is still acceptable from a human dignity point of view. So I think we really have a challenge ahead of us to look at everything related, all the human rights, and probably interpret them from very profound principles. And this is, i think, also how I understand the quote that I used from Pope Francis. We really have to go back to these questions, and of course It's very complicated.
Thank you.
Thank you very much, Ambassador. With that, thank you. Also, please join me in thanking the panel. And thank you also for the audience to staying in such a great number on a Friday afternoon. Thank you very much. And I'm handing over to Wafa, our Digital Tech and Human Rights Section Chief, for closing remarks. Please give her your full attention for the closing.
Hello, everybody. It is the end of a day, and I had only a few things to say just to tell you about a few next steps that we're doing. My name is Wafaa Bin-Hussein. I'm Chief of Section at the UN Human Rights Office, Section for Digital Technology and Human Rights. Thank you all for coming here today, distinguished participants and excellencies. As we come to the close of today's discussion, I would like to extend my sincere thanks to all of our panelists for their thoughtful contributions and to all of the participants and for your active engagement throughout the afternoon, especially once again on a Friday. The richness of today's discussion reflects both the complexity and the importance of the issues before us. Rather than attempting to summarize all the excellent interventions, let me simply note that our exchanges have reinforced the importance of ensuring that the development, deployment, and governance of digital technologies remain firmly grounded in human rights. Today's discussion forms part of a broader process. The perspectives, lessons learned, and recommendations shared during this intersessional meeting will inform a summary report to be prepared for consideration for the Human Rights Council in 2027. I will also add on the note that Meredith from the Business and Human Rights Center just said earlier about reclaiming the term innovation. I would go a little bit further and say that we should also reclaim the term safety, because it is human rights defenders that have worked tirelessly for almost a decade now, working on safety and AI, and seeing how human rights values can best inform how to go about it. I'd also insist that we carry human rights into spaces that we're not typically part of, like security and technical standards. We need to be able to work across sectors. Before we conclude, I would also like to highlight a number of opportunities to continue these conversations in the coming months. Just next week, OHCHR's BTEC project will be organizing a series of events on the sidelines of the UN Forum on Business and Human Rights, and our office will conclude a series of five regional consultations on children's rights in the digital environment, convened pursuant to Human Rights Council Resolution 56/6. This process will provide an important opportunity to ensure that children's voices and experiences help shape the digital policies and governance frameworks that affect them, and my office will present a report at the Human Rights Council at the 64th session that will inform an interactive dialogue among states. And last, but certainly not least, I would like to draw your attention to the upcoming Global Alliance for Human Rights Forum in Geneva on the 16th of December. Among its dedicated deep dive conversations will be a session on agentic AI and human rights, providing a timely opportunity to continue examining how rapidly evolving AI systems can be governed and deployed in ways that respect and advance human rights. We hope you'll be able to join us for that conversation. Looking further ahead, of course, there are a number of important global and multilateral processes, including the second Global Dialogue on AI Governance in May, followed by the WSIS Forum and the AI for Good Global Summit in June, and the Global AI Summit here in Geneva. Together, all of these initiatives offer valuable opportunities to advance human rights-centered approaches to AI governance. I would also like to take this opportunity to remind participants that OHCHR stands ready to support these efforts through its advisory services. Whether states, companies, or civil society organizations are seeking guidance on integrating human rights into digital technology governance, strengthening policy or regulatory frameworks, or implementing human rights due diligence approaches, our office does remain available and ready to provide support and expertise. We hope many of you will continue to engage with these and other related initiatives. Finally, I would like to reiterate our appreciation to the Republic of Korea and the rest of the core group, to our panelists for their valuable contributions, and to all of the participants for your commitment to ensuring that digital technologies serve human dignity, equality, and human rights. And to our interpreters for their great support, thank you. I wish you a very good evening, and I look forward to continuing these conversations in the months ahead. Have a nice weekend.