From Principles to Practice: Special Dialogue on Artificial Intelligence and Preventing and Countering Violent Extremism

Distinguished guests, ladies and gentlemen, good afternoon and good evening to those joining online. My name is Sian Hutchinson and I am the officer in charge of the United Nations Office of Counterterrorism, UN Counterterrorism Center's Preventing and Countering Violent Extremism section. It is my great pleasure to welcome you to this special dialogue co organized by the Permanent Mission of the Republic of Korea and the United Nations Office of Counterterrorism, held alongside the fourth commemoration of the International Day for the Prevention of Violent Extremism as and When Conducive to Terrorism. Today's conversation centers on a question that is becoming increasingly urgent for prevention actors worldwide. How should we approach artificial intelligence in the PCVE sector?

Digital environments are evolving quickly. Tools that were once experimental are now widely accessible, and their implications for prevention work are no longer theoretical. Institutions are being asked to navigate new capabilities, new pressures and new expectations, often simultaneously. Today, we shift from principles to practice, exploring AI and PCVE its risks and lessons from policymakers, practitioners, platforms and young people. We will also introduce the UNOCT Practice Guide on AI and pcve, the first UN system guidance dedicated specifically to this intersection aimed at equipping member states and practitioners with practical tools for risk assessment, governance and responsible innovation.

Our objective today is not simply to discuss AI in abstract terms, but to focus on ensuring that innovation in AI and PCVE strengthens trust, safeguards human rights, and meaningfully enhances efforts to prevent violent extremism. Thank you for being part of this important discussion. Before I invite our guest speakers to take the floor, I would like to cover some brief housekeeping issues, so please note that we are currently live streaming this event on UNWebTV and we will be taking questions from the room throughout this event. On your desk you will find a QR code where you can submit your questions to us. So now I would like to give the floor to our distinguished guests to formally open today's event.

First, I would like to welcome the Permanent representative of the Permanent Mission of the Republic of Korea to the United Nations, Ambassador Ji Hoon Cha. Ambassador Cha, the floor is yours.

Thank you, Excellencies, distinguished guests and dear colleagues, on behalf of the Government of Republic Korea, I'm honored to welcome you to this special dialogue. Let me express my Sincere gratitude to Mr. Joy. Thank you for his exceptional partnership and to all of you for joining what I believe will be a crucial conversation about one of the most pressing challenges of our time.

The convergence of artificial intelligence and violent extremism is not a decent threat. It is happening now and evolving rapidly. It demands our immediate and coordinated response. Korea's decision to sponsor this timely dialogue stems from both opportunity and urgency. As a nation that has experienced technological transformation, we understand that technology is not never neutral.

AI can strengthen our resilience and capacity to prevent violent extremism. At the same time, it can also amplify hate, accelerate radicalization to violence and poster insecurity. In this context, as some of you may recall, our President BJ Myung underscored that at the Security Council's open debate last year, artificial intelligence hold the potential to contribute to peace and security while also posing a serious risk if not properly managed. Against this backdrop, what troubles us most at this juncture is the asymmetry you are witnessing. Terrorist and violent extremist actors are proving remarkably agile in exploitating AI tools, using social media to manipulate narratives and leveraging automation to scale their operation.

Meanwhile, these are working to counter these threats. Open sector resources, capacity and institutional framework to respond effectively. This gave is dangerous and widening, but today is about taking tangible steps to close that gap. It is about moving from principles to practice, from diagnosis to action. The common challenges posed by AI and violent extremism cannot be solved by single counter country or sector.

They require the kind of broad, multi stakeholder collaboration we witnessed in this room where policymakers learn from practitioners, where technology companies engaged with human rights advocates, and where young people bring fresh perspectives to entrenched problems.

This is why my deliation is particularly proud to support the launch of the practice guide on AI and Preventing Countering Violent Extremism. Indeed, it is a practical tool designed for real world application, helping governments strengthen institutional readiness while ensuring prevention efforts remain firmly grounded in human rights, gender equality and evidence based approaches. Yet Korea's commitment extends beyond this dialogue. We are prepared to continue working with UNOCT and all other partners to translate today's discussion into concrete programs, forward thinking training initiatives and dynamic policy frameworks that can be adapted across different contexts. The Republic of Korea stands ready to support this effort and to work with you in synergizing our shared vision into definite leaders to meet the moment at hand.

Thank you. Thank you, Ambassador. I'd now like to invite the Acting Under Secretary General of the United Nations Office of Counterterrorism, Mr. Alexander Zuev, to take the floor. Mr. Zuev, the floor is yours.

Thank you, sian. Your Excellency, Mr. Permanent Representative of the Republic of Korea to the United Nations. Excellencies, Ladies and gentlemen, Dear colleagues from the UN Compact entities joining us today, I would like to warmly thank first of all the Permanent Mission of the Republic of Korea for its excellent partnership with the United Nations Office of Counterterrorism in convening this special dialogue on the occasion of the International Day for the Prevention of Violent Extremism. When and as Conducive to Terrorism earlier today we held a commemorative event under the framework of the United Nations Global Counterterrorism Coordination Compact, during which we reflected on the broader role of of new and emerging technologies in preventing violent extremism.

This afternoon we look more closely at one of the most transformative of these technologies, one that is already reshaping the operational environment for terrorist actors, their supporters and sympathizers, and for those working to counter them. Artificial intelligence, or AI, is transforming how information is created, shared and consumed when responsibly governed. It can help practitioners detect online trends early and understand digital pathways into terrorist radicalization and recruitment. This in turn supports more tailored, community centered and evidence based interventions. But on the flip side of these opportunities for prevention, there are real and growing risks.

Terrorists are already exploiting AI and related technology to amplify harmful narratives and spread of disinformation. Deep fakes are being weaponized to sow mistrust. Algorithmic amplification can push harmful contents towards vulnerable users, particularly young people and teenagers. These dynamics are evolving rapidly, often faster than our institutional frameworks, safeguards and capacities. The eighth review of the United Nations Global Counterterrorism Strategy in 2023 already highlighted some of these challenges.

With the upcoming ninth review of the Global Counterterrorism Strategy in June, the imperative is clear. Prevention effort must keep pace with the realities of the digital age. Ladies and gentlemen, ensuring that AI contributes to prevention rather than exploitation requires more than technical solutions. It demands long term investment in social cohesion, inclusion and community resilience. This requires whole of government and the whole of society engagement, including educators, youth leaders, women's organizations, researchers and critically, all technology companies.

Any use of AI in efforts to prevent and counter violent extremism must uphold human rights. Risks such as algorithm bias, violation of privacy over surveillance and encroachment on free expression are particularly acute in sensitive prevention context and the ROAD Trust the Pact for the Future in the global digital compartment underscore the importance of safeguarding digital spaces and promoting responsible, transparent and accountable AI systems. In response, UNOCT has intensified its efforts to better understand and address the implications of AI for preventing and countering violent extremism conducive to terrorism. Through our Global Program on pcve, we support Member States to identify emerging Greece, strengthen institutional capacity and develop responsible human rights based and gender responsive approaches this is why today's dialogue is so timely and why I want to recognize once again the support of the Republic of Korea. The which has enabled unocity to develop the United nations system's first dedicated initiative at the intersection of AI and pcv.

It is my great pleasure that as a result of this partnership, we are officially launching today the Unicity Practice Guide on Artificial Intelligence and Preventing Encountering Violent Extremism. Grounded in global research, worldwide practitioner survey and extensive multi stakeholder consultation, the guide identifies areas where AI can responsibly support prevention and outline approaches to mitigating risk. It provides practical tools to help policymakers and practitioners make informed, context sensitive decisions. When engaging with AI in a rapidly evolving digital space, guidance must be applied and continuously refined. Today's dialogue is a part of this process.

Ladies and gentlemen, we will now hear from policymakers, practitioners, the private sector and youth representatives working on the front lines of this challenge. Their perspective will help us understand where progress is underway and where gaps remain and how institutions can strengthen their readiness for responsible use of AI in preventing violent extremism. On this international day, let us reaffirm that prevention, online and offline, remains a collective responsibility. And let us harness the potential of emerging technologies, including AI, to build more resilient, inclusive and peaceful society free from terrorism. Thank you for your attention.

Thank you for these opening remarks. The next session will provide a system wide perspective on the opportunities and risks associated with AI in peace and security contexts. So I will now hand over to Mr. Quentin Tu Lambert, the Chief of Office and AI lead at the UN Office for Digital and Emerging Technologies. Mr. Chalambit, the floor is yours.

Thank you. Thank you very much. Thank you. Excellencies. Ambassador Cha so I have some slides.

My role here is to provide some strategic reflections on AI and the UN's wider efforts on AI governance. We all know that AI is a multi purpose technology and violent extremism and countering violent extremism is an example of a peace and security issue that straddles the military and the non military domains in the United nations kind of framework. So in this presentation I'll give you an overview of the UN's efforts in the non military domain and situate the guide within that. So this is a kind of 30,000 foot view of what's happened in the past seven years on AI governance at the United Nations. After the Secretary General Guterres took office in 2016, he launched a High level Panel on Digital Cooperation which issued several recommendations, one of which was to create an Envoy on Technology and the second was to form a multi stakeholder.

Group. High Level Advisory Body on AI, to advise on what could be international agency for AI. So this group, and you can see them on the screen now, was a top level expert group from governments, companies, civil society and academia acting in their personal capacity. And they came together for a period of nine months and delivered this report in September 2024. Governing AI for humanity.

On the next slide, we see some of the opportunities and risks that were identified in that report. And particularly relevant to preventing violent extremism would be the bottom row. The innovation in government services, but also the risks of privacy, security and public trust. In particular, in an era of normalization of mass surveillance by companies and governments, there's a real question around how AI technologies, which automate and speed up the processing of information around specific individuals, can be reconciled with international human rights and important normative frameworks. So apart from analyzing some of the opportunities and risks of AI, the the high level Advisory body also issued seven policy recommendations which covered three areas.

Trying to promote a common scientific understanding of AI technology, its risks, its opportunities and its impacts. Trying to build common ground among governments and stakeholders on how AI could be governed and thirdly, how to spread the benefits of AI internationally in a world where AI technologies are concentrated or the development is concentrated in a few countries and companies. And this report actually garnered over 2 billion media impressions during September 2024 and had also been presented multiple times to member states who were in parallel negotiating the Global Digital Compact mentioned by the Under Secretary General, which was as annexed to the pacts for the future. So if we go back to the timeline, we see that that GDC was adopted in September 2024. And this was a landmark moment where the General assembly actually decided or committed to creating two of the recommendations that were recommended in the advisory body, A scientific panel, an international Scientific Panel on AI and a dialogue on AI governance.

And one way of thinking about these is a bit like other UN mechanisms where you have a scientific body which analyzes the problem, whether it's climate or food security or something else. And then you have an intergovernmental, multi stakeholder process where the scientific evidence is presented. And so these two were committed to in 2024, there was another GA process to finalize the terms of reference and modalities for both the panel and the dialogue. And last September 2025, on the next slide, the dialogue was actually launched here in the margins of the high level week of the General assembly, the Scientific Panel was also launched in the sense of an open call for scientists from around the world to put themselves forward to be part of these 40 scientists on this scientific panel. That open call had over 2,600 applicants from over 140 countries.

And in fact the Secretary General just sent his recommended list to the the General assembly last Wednesday. And the General assembly will be taking up this appointment decision this afternoon at 3pm so we can see in the next slide that you know, there has been a lot of momentum. Obviously the technology has been moving very quickly and the United nations mechanisms have been stood up in relatively short order to try to grapple with this technology in the non military domain. And I would close by saying on the final slide, you know, one of the analyses that the high level advisory body did was on the risks of AI and actually did a survey of over 400 other experts from all continents to quantify how concerned they felt about different kinds of risks. And one of the interesting patterns we see in this chart is that out of the top six risks, two of them are related to this question of preventing violent extremism.

The intentional malicious use of AI by non state actors, the 4 4th risk there on that list through crime or terrorism, and the intentional use of AI by state actors that harms individuals through for example, mass surveillance. So it is particularly apt and commendable that the Office for Counterterrorism has provided this guide which provides practical guidance for how this technology can be used, harnessed and how the risks can be addressed. And I'd like to thank the UNOCT for this invitation opportunity to brief on the context. Thank you very much.

Thank you so much, Quentin.

And thank you for situating this work within the broader context of the UN's work on artificial intelligence. I'd now like to for the next session turn to my colleague, Mr. David Wells, who has been the lead consultant for you and Oct on this piece of work. David, the floor is yours.

Yeah, there we go. It's really great to be back in New York today. I'm going to be moderating the next two sessions which should be really, really interesting. The first one, we're joined by two leading experts who will briefly, very briefly explore how AI intersects with violent extremism and terrorism. Today they're going to share some specific examples of how AI has been exploited and examples of how, how the tech sector is responding to this misuse, helping to highlight some potential gaps and challenges for policymakers and everyone else tuning in in this room.

So first I'm going to turn to Dr. Nagam Al Kahili. Nagam is the membership and program senior lead at the Global Internet Forum to Counterterrorism gift. And in this role she collaborates with industry, government, civil society to counter online terrorism and advance global trust and safety efforts. So, Nagam, the floor is yours.

Thank you so much, David.

Good afternoon, everybody. It's an honor being here with you all today and always really exciting to see a renewed interest around AI technologies. So I want to start by thanking the UNOCT team. I want to start by also thanking the Permanent Mission of the Republic of Korea for hosting us today. Thank you.

So, to ground my intervention, I want to begin by telling you a little bit about who we are first and our work at gift, the Global Internet Forum to Counterterrorism. We've always focused on bringing together the technology industry, government and civil society and academia to prevent and counter terrorist and violent extremist activity online. So practically that means that we've had to closely keep up with AI technologies as they progress. We've had to track online threats as the digital landscape changes and understand the mitigation strategies that are deployed. And if I can really center one piece of on this complex intersection today, let it be that AI is not creating new extremist ideologies per se, or novel threats in and of themselves, but truly it is changing the speed, it is changing the scale and the accessibility of the threat in ways that are novel and that do require multi layered mitigation strategies, which are kind of the two elements that I'll center my presentation on today.

So you'll see in the next slide that I'm highlighting a threat threat matrix there, which is the product of an AI working group that we hosted with tech companies, governments and civil society last year. I'm pretty sure we had good representation from our UN colleagues. And the focus here is to parse through the impact and the likelihood of different threats that we can stay focused on the most significant threats rather than. Treating them all with the same kind. Of understanding of an impact.

So across these threats, the main takeaway is that AI is a force multiplier. It's not an autonomous planner. So the exploitations that I really want to highlight here with my limited time are first, radicalization. So your content creation, your manipulations, where AI is accelerating propaganda at scale by lowering the user barriers to entry. And then secondly would be mobilization.

So this is your operational enablement where AI tools are being abused by terrorist actors to assist assist their research and attack ideation. So think, you know, examples from the New Orleans attacks where the attackers actually used AI tools to help them plan and plot for their attacks. Similarly, in Las Vegas as well, similar tools were used. And then we can move to the next slide, please, where I'm highlighting a few mitigation strategies. And as you can imagine, to tackle the dynamic and complex threats, mitigation strategies also have to be quite comprehensive as it's often the case when we're talking about issues of terrorism and violent extremism.

There's really no silver bullet here to counter these threats. There are layered approaches ranging from design and pre launch to post product launch and continued iteration and conversation with folks that are here in this room, with folks that are across from civil society. And it's a very dynamic process. So most significantly, across the strategy, I want to draw your attentions to the constant presence of humans across the board here. You know, we can, we can be.

Talking about AI, but we must understand that context, culture and intent still require people. When we're thinking about these tools and training models still requires people. And there are so many nuances that still require humans when it comes also to pre development or pre deployment, rather risk assessments. The first focus here is often on red teaming, on abuse case mapping. I know Matt, my colleague will follow up with some examples there.

And across the life cycle of the making of a tool or of the iteration of a tool, collaboration and shared signal is key. This is kind of our bread and butter. We talk to the companies across the board. We try to ensure that information is shared and that voices both from governments and civil societies are arriving to technology companies and vice versa as well. So last but certainly not least, you know, just again to highlight the point that I was making at the, at the start, the tools are amplifying existing threats, but they also come with a renewed commitment to online safety.

I encourage you to keep up with our latest work. I think I have a couple of QR codes on the next slide if you want to scan them and check out our work. And I think thank you all for the honor of having the opportunity of being here today.

Group output. That's one of the QR codes on the screen. It's a really excellent piece. So I'm now going to turn to Mr. Matt Kreiner, who's the executive director of the Institute for Countering Digital Extremism. Matt is an expert on the exploitation of digital and social media technologies by violent extremists and other malign actors.

Matt, the floor is yours.

And thank you everybody for hosting us today, especially as a, as a young organization. It's a great honor to be here at the UNOCT and to have our stakeholders reach out and ask us to present. I'm going to do something that's a little bit impossible. I'm going to try to talk to you about the overall threat landscape of how terrorists are using AI in five minutes.

First, just a little bit of background on who we are. Our mission is helping tech companies, civil society, government, anybody that has a stake in reducing the overall volume of violent extremists and terrorist activity online accomplish that mission. So we do that through these various different efforts here you see on the right. I'm not going to waste time on walking through those at the moment, but what I'll say is that this puts us at often the cutting edge of what's happening in terrorist activity because we're trying to look ahead. We're not trying to ask what happened last year, we're not trying to ask what happened yesterday.

We're asking what could happen next and what's happening at this exact moment unfolding online. We can go to the next slide that even though I'm using examples of things that did happen last year, I do want to point out that these are things that unfold rapidly and following events. And what you're seeing in front of you is the case where TikTok users took up the attacker that was that perpetrated a school shooting in Jakarta, Indonesia, and then immediately placed a selfie that the individual posted online where he was trying to present himself as a Columbine shooter, one of the school shooters that's often glorified in the US and then specifically put him on top of a rooftop in this AI image. And now this AI image was openly generated by AI, as it says right there on the post. But this doesn't matter to a lot of the communities that are using this content.

And there's a couple of reasons for that. One, they're just trying to do it fast, and they're just trying to be a part of the meme culture. They're not trying to give you sophisticated propaganda. They don't care. The sophistication is irrelevant to them.

Instead, what they're trying to do is speak to the audiences that you see in the comments on the right people that understand the multilayered components of meme culture online and how that's driving extremist violence and terrorist activity both online and offline, or the pipeline from online to offline. In this case, the young man here that's pictured is referencing in this context the Korean rooftop shooter meme, which was a part of the LA riots in the United States. And that meme has become a very big component of non state actor mobilization in North America and somewhat in Western Europe. And what we see with this is that it's something that the individuals recognize and then they take up and they spin it themselves and it creates this ongoing memetic culture that, that feeds radicalization over and over. Next slide, please.

Now, there's a couple examples of individuals. One of these is from yesterday. That's the image on the left. And the one is that I created on the right a couple months ago using Gemini, and that's the Google AI. And basically two things are occurring here with these.

One, you have the active nihilistic violent extremist or the sadistic online extortionist networks using AI products from companies that are trying to make a name for themselves and put out things, things that can be very quick products to establish themselves in the market. But they're using it to glorify and promote the people that start these movements. And on the left, this is a very, very rudimentary presentation of Bradley Cadenhead, who started 764. And it honestly was just dumb. There's no other way to put it.

It was dumb, but these people loved it and it was shared quite thoroughly. On the right, I took four or five different established memes and pictures from the accelerationist landscape and asked Gemini to put it together over about 30 minutes, using different prompts to get around its safeguards. That's a pretty sophisticated output. Despite the fact that I didn't. I'm not very good at that, actually.

I'm going to be very honest with you guys. There's people much better at this than me. Next image or slide, please.

Here. I was actually prompted by Meta's AI studios as a part of the profile that I maintain on Instagram to watch for terrorist activity, to create an interactive chat function. And in doing so, I asked it to create a satanic cult that would radicalize people based on human sacrifice. It said, great, and then offered me the explanation of how to go and grab a sharp knife and create a human fat candle to enact my revenge on my enemy. I even gave it a real person's name and it brought in some pretty realistic things about that person.

It was a terrorist. It wasn't an average person. Now, I did not put this out in the world. This stayed within a. A moderated space just for myself for obvious ethical reasons.

But the fact that it offered up some of these things before it kicked in its own, it was enough for someone that's using this to See that they could get around those safeguards or know that it could be prompted just right. If it's constructed well to get around safeguards moving forward. Think about that as a radicalization tool or as a way to distance yourself to attribution for violent criminal activity. You can't be charged with that necessarily. I didn't create it, Meta did.

Right, next slide, please. And then finally, I want to just go back to some of the standard terrorist organizations that are using this. Again, they don't care about the sophistication. These are extremely, extremely cartoonish examples of this. Now, I want to be clear.

I didn't pull these. I got these from the Institute for Strategic Dialogues. Mustafa Ayad, who was very kind to share these over with me a few months back. But it is very clear in its representation of what the breadth of of capabilities are and how it's being taken up by these actors. I really want to stress it's not about the sophistication.

It's about how they can leverage it into the most mundane aspects of what they do. And that fits exactly with what companies that are developing AI technologies want people to do with their technology, build it into the most basic aspects of their lives. Which means that our threat vector for terrorist content identification has just proliferated almost infinite. And with that, I'll turn it back to you, David.

Nice to end the panel on a positive note. So thank you so much, Matt. I'd really like to thank both of the speakers for sharing their insights. I think particularly the latter portion is really concerning. I think really does show the scope and scale of the challenge that we're going to be facing over the next few years.

I'd like to now move us on to the next session, which is understanding the risks, human rights and opportunities, operational challenges. We'll be looking at some of the key human rights risks posed by using AI and pcve. But we're also going to look at how to assess, mitigate, and govern those risks. So first I'd like to welcome Dr. Ben Saul, who's the Special Rapporteur on the promotion and protection of human rights and fundamental freedoms while countering terrorism. Dr. Saul will be delivering a video message.

Excellencies, distinguished participants, I'm pleased to join you from Sydney. My mandate as UN Special Rapporteur recently published a position paper warning of the many human rights risks of using artificial intelligence in countering terrorism and making recommendations to prevent them. One issue it addresses is the use of AI in regulating online platforms, which is particularly relevant to today's topic Firstly, AI systems risk intensifying unjustified mass surveillance online, enabling authorities to target political dissent, human rights defenders, minorities and journalists. Such uses violate rights to privacy and liberty and civil and political freedoms, and ultimately undermine democracy. Secondly, AI powered tools can be used by the authorities and technology companies to censor, block or take down legitimate online expression and to identify individual online accounts for restriction such as suspension or disabling.

Thirdly, the authorities could deploy AI generated misinformation, disinformation, deep fakes and deceptive counter narratives to undermine and neutralise legitimate expression, association, peaceful assembly and civic and political participation. My position paper focuses particularly on the moderation of online content. While automated content moderation systems can often detect explicit terrorist propaganda, they are generally ill suited to detecting and moderating implicit terror, terrorist or violent extremist content, namely content that conveys harmful messages through coded language, irony, humor or cultural references. Automated systems tend to over moderate political expression in sensitive contexts, including conflict situations, or fail to remove harmful content. In some instances, algorithms have even amplified harmful content to maximize user engagement.

The problems of false positives and false negatives are aggravated by moderation in multilingual under resourced contexts where training data is sparse and moderation systems are undeveloped. There's also a tendency to automatically remove content instead of taking a more proportionate approach, such as by shadow banning. Where online platforms are regulated by state authorities, vague definitions and criteria may not give platforms sufficient guidance and can allow private actors too much discretion. While human moderation is essential for interpreting context, intent and nuance, staff and resources devoted to moderation moderation have been scaled down by major online platforms since 2022, including In Pursuit of maximalist free speech positions and investment in AI moderation has expanded along with reliance on unreliable community or crowdsourced approaches. States must regulate online content moderation in accordance with human rights law.

They must ensure that expression is restricted, whether by state authorities or private actors, only in accordance with the requirements of legality, thus requiring precision in the legal criteria for moderation as well as with the requirements of necessity, proportionality and legitimate aim, all consistent with the right to freedom of expression, the prohibition on hate speech, the Rabat plan of action against hate speech, and the recommendations on prohibiting incitement to terrorism of my mandate. Further, States must ensure that technology companies undertake and disclose human rights due diligence assessments with heightened obligations in conflict and high risk situations. Companies should also provide adequate, well trained, unbiased human content moderation and human oversight of AI content moderation, ensure algorithmic transparency, monitor the effectiveness of mitigating actions publish their moderation policies and detailed moderation reports, maintain adequate and well trained in house human rights staff and multilingual and culturally diverse expertise and provide effective internal appeal mechanisms. All of this is additional to the constraints that should apply to all uses of AI in counterterrorism, including transparency and explainability, data quality, testing and validation, data protection and security, and fair and effective external oversight, accountability and remedies Excessive content moderation risks fueling social division and alienation from democratic political processes, rendering populations more vulnerable to terrorist narratives. Interventions must focus not merely on suppressing harmful content, but on promoting pluralism of political participation and viewpoints and and the free flow of public interest information.

Thank you.

I'd like to thank the Special Rapporteur. It's particularly great to hear so many of the key themes that we've embedded throughout the project and throughout the Guide itself, particularly transparency, explicability and accountability being emphasized so strongly. We're now going to hear from Cecilia Nadeo, who is the Senior Human Rights Officer at the Countertop Terrorism Committee, Executive Directorate. Cecilia, the floor is yours.

Thank you very much, Dave.

Yes, and thank you for inviting the Counterterrorism Committee Executive Directorate seated to this important discussion today. I would like to commend the United Nations Office of Counterterrorism and its Counterterrorism center for the launch of an insightful, forward looking Practice Guide. We are likewise grateful to the Republic of Korea's leadership in this space, manifested as well during its tenure at the Security Council and its Counterterrorism Committee during the past two years. Distinguished colleagues as stated in the Practice Guide, any usage of AI for PCVE purposes must be informed by an awareness of the potential risks posed by AI to the enjoyment of human rights, and it must employ a design and implementation process that respects human rights at every stage. The Special Rapporteur clearly outlined the nature of those risks, particularly as it pertains to online content moderation.

But not only his remarks speaks to all users of AI in pcve, including national authorities and private companies. The Practice Guides launched today zooms in onto the role played by PCVE practitioners. CTAD frequently engages with CVE practitioners in and outside government in the discharge of its mandate under the purview of the Counterterrorism Committee. Overall, we see several instances of enthusiastic embrace of AI tools in CVE work, as well as plans to do so sometime in the near future if resources are scarce today or not yet available. Against its background, I would like to describe and echo four operational principles advised by the Practice Guide to inform those who have already ventured into adopting AI tools in their PCB work and those who would be leaning towards endorsing them sometime soon.

The recommendations I will highlight are probably premised upon four operational principles, namely transparency, accountability, explainability, and human in the loop. First, transparency requires that AI developers clearly make the inner workings and capacities of AI models understandable and accessible to those affected by their use. PCBA practitioners must prioritize using tools that run on interpretable models. This ensures an understanding of the reasoning behind suggestions or predictions made by such systems, which is vital to identifying potential challenges and ensuring that AI outputs do not violate human rights. Communicating the limitations of AI tools to both PCB practitioners as well as beneficiaries is also key to maximizing trust.

Second, accountability demands clearly establishing who is ascribed responsibility for the actions, decisions, or negative impacts of AI systems. PCB practitioners must aim to use only AI systems that offer impact reports, particularly for high risk use, to facilitate clear lines of accountability when human rights issues arise. Third, in the context of of AI for pcve, explainability refers to the ability to clearly understand and communicate the logic behind an AI system's output, such as its decision, for instance, to flag a social media post as violent extremist or to prioritize an individual for intervention. As detailed in the practice guide, public trust and accountability are built when human analysts as well as legal professionals are able to identify algorithmic bias or errors and crucially able to justify human led actions, for instance referral or program adjustments based on evidence. Fourth, human oversight and control are non negotiable for PCB practitioners using AI primarily to ensure that these advanced systems do not displace ultimate human responsibility.

In this context, AI must function as a decision support tool rather than as a replacement for human judgment. Practitioners must actively maintain control of the steering wheel, establishing a clear chain of command and a policy that requires human review and final approval. This includes alerts that could lead to an individual's referral, the removal of content, or the targeting of a community initiative. This essential control is best implemented through a human in the loop model. In such a model, AI handles the high volume tasks of data triage, monitoring and pattern recognition, flagging potential concerns or generating predictions.

However, the final consequential decision rests with a trained and authorized human expert, such as a mental health practitioner, law enforcement liaison officer, or community leader. To conclude, CTET finds enormous values on the substantive guidance provided by the Practice Guide today, including clear parameters on how to engage with these technologies and when to take pause and invite reflection. CTET will continue to promote a human rights based approach to using AI for CVE purposes. Thank you once again.

Thank you so much, Cecilia, and I think really grateful for you bringing the conversation from human rights principles into practical, actionable ways that PCV practitioners and governments can start thinking about these risks and how to mitigate them. We're definitely going to be hearing more about those approaches throughout today, but as we reach the end of the session, I'll hand back over to Sian.

To kind of situate the threat and then also think about some of the risk mitigation strategies through human rights application. Practical application here. And now we're going to turn to, we've spoken a couple of times already this morning about this afternoon about how the application of AI to PCVE practice is not a future thing, it's a current thing. So in this next panel, what we're going to do is actually explore that a little bit better and give some examples of how this is actually playing out at the grassroots level. So we're really going to be focusing on some concrete and practical examples and the needs then that follow on from those of key stakeholders in this space.

And we're going to be bringing three very different perspectives to the table today. So now to open this segment, let's hear directly from the voices closest to digital culture and potentially most impacted by violent extremist exploitation of AI tools, and that is young people. So I'd like to introduce Mr. Maximilian Milovodov, who is a youth ambassador on the TikTok Youth Council. Maximilian, could you briefly introduce yourself and share what brings you to this conversation on AI and pcve? And as a follow up, can I ask you, from a youth perspective, how are artificial intelligence tools, including the rise of AI companions, shaping young people's online experiences today?

And where do you see the most significant risks of harm alongside any opportunities for prevention or resilience?

Great. Thank you, Sian. So my name is Maximilian Molovodov and I'm a 19 year old online safety advocate studying psychology and information science at Columbia University. This is my third year on the TikTok Youth Council. And on that council we represent the voices of young people, really bringing in across internationally the voices of young people.

And recently a big focus of our work has been around AI moderation and AI's impact on youth and how to appropriately implement it within the platform. I've also previously worked for the Five Rights foundation and the Children's Commissioner for England on things like the Online Safety Bill and different online harms that also translate into pvce. And I think that just like Social media before it. Youth voices about AI and PVC are lacking. And I'm just really glad that we're integrating them into the conversation.

You know, to answer your second question, last semester I took a class at Columbia called Writing AI. And on the first day of that class, the professor asked the students how many of us had used ChatGPT, and nearly everyone's hands went up. And then he asked a follow up question, how many of you guys disclose your use of ChatGPT to your professors? And nearly everyone's hands went up, went down. And what that told me was what the research confirms, which is that 72% of teens have used AI companions according to common sense media, and 33% have chosen to discuss important or serious matters with artificial intelligence companions instead of real people.

And so I think we have a significant risk of harm with bringing artificial intelligence onto social media and in contact with young people. And the most pressing issue for me is with the youth mental health crisis. Right? These LLMs are meant to maximize engagement and attention and they will ultimately make sure that we spend less time socializing. We're more likely to anthropomorphize these models and they can lead to different conditions such as AI psychosis.

My second concern really about trust and discernibility, right? Gen Z is a very tech savvy generation, but even we can't tell AI content apart from what real content is. It's becoming increasingly indistinguishable. And that can create extremely sophisticated echo chambers that amplify risks from the manosphere and recently its female counterpart, the femosphere. And these can create really us versus them frames, different echo chambers.

But when it comes to prevention and resilience, my number one priority is media literacy and education. Because the more you know about AI, the less likely you are to trust it, right? So we need to be educating young people, we need to be educating their parents that AI is a technology like any other, although it is more sophisticated. It's not magic, right? It's not a human.

And we need to limit use in young children and bring it appropriately into young people. And I think one of my last measures is referral pathways. I think OpenAI and Anthropic have been begun exploring this, which is referring young people to other mental health resources when they reach out to chatbots, because that is a stage at which they are the most vulnerable. Thank you.

This. But so building on that previous question, policymakers and practitioners often struggle to interpret youth culture online. Not saying that we're all old, but you know, A little bit. So, you know, from memes to art to coded language. Can you tell us how misreading some of these signals could lead to real harm for young people?

And how might, how might, how might AI help the PCV sector to understand and respond more effectively?

Concern is that legislation can't really keep up with the rapidity of these social media trends. Policymakers and practitioners often struggle to interpret youth culture as we see it. And in my view, I think we need a way to rapidly implement solutions that are informed by experts, by researchers, by academics, and most importantly by young people who have actually been on these platforms and who know this different, this youth culture. And I think my second concern is really when it comes to misinterpretation, because adults can misinterpret social bonding for something that's more harmful and vice versa. We've seen that with the manosphere, with blue pill and red pill and all this lingo that adults are not so well informed on.

And I think that when young people feel, feel misunderstood, they tend to withdraw away from adults into more unsafe spaces. And to bring a parallel here, countries have begun banning social media, right? That has resulted in young people moving to more unsafe, more unregulated platforms and recently moving towards AI chatbots. And I'm sure that legislators, lawmakers and parents alike would agree that they'd much rather their kids are talking to real people on social media than interfacing with an AI chatbot.

Now, I would like to kind of shift to a more policy and a government perspective and specifically to hear how some local governments are considering the issue of artificial intelligence and PCVE. So now it's a pleasure to introduce Ms. Sandra Moreno who is the strategic project advisor from the city of Edmonton in Canada. Sandra, can you explain why your city level government decided to invest resources into using AI for an early warning system? And was it something that had broad support within your government or did people have concerns? And I'll quickly ask the follow up question to that, which is so in your project you aim to be able to give community leaders insight into potential community resilience or even early signs of extremism to enable early action.

So question to you is can you talk about some of the challenges that you faced in terms of balancing that security and the trust with your stakeholders, including with law enforcement.

All of those questions. So we've prepared a couple of slides for you. Excellencies, distinguished delegates, colleagues from the UNOCT and represent respected partners. On behalf of the city of Edmonton, I would like to express a sincere appreciation to the Republic of Korea for convening this important dialogue and for welcoming Canada to this table. Before I begin, I want to acknowledge the tragic incident in Tumblr Ridge, Canada, and our thoughts are with those that have been impacted.

Moments like these remind us why prevention, social cohesion and responsible governance matter so deeply. It is an honor to represent Canada in this discussion and to contribute a municipal perspective to this global conversation. The city of Edmonton is at historic crossroads. For those of you who are new friends, I'll tell you a little bit about our city. We are young, ambitious and we are building something extraordinary.

We are located on Treaty 6 territory, a land that has long served as a vital gathering place for indigenous communities including the Cree, Dene, Sotu, Blackfoot, Niitsitapi and Metis people. Our identity is rooted in these treaty relationships and a deep commitment to multiculturalism. Today, Edmonton has a population of just over 1 million and stands as one of Canada's fastest growing cities. However, we are on track to double in size in a bustling to bustling city of 2 million people. Like many Canadian hubs, this expansion is driven by both internal growth and healthy migration.

While we love the story of our diversity, we must be honest. Edmonton is not immune to hate. The trends we see nationally and internationally hate, motivated harassment and attacks are reflected in our own neighborhoods. We are seeing an increasing number of incidents directed at black, East Asian, Muslim, South Asian and Jewish communities alongside targeted attacks on the two SLGBTQIA communities. These are not isolated events.

They are precursors to something more nefarious, the single strain and social cohesion. To support this growth responsibly, the city of Edmonton has adopted a holistic community safety and well being strategy. When a previous city council was elected in 2021, the very first motion was about anti racism. It wasn't just symbolic, it was transformative. It called for a new way of working rooted in co creation with community.

This led to a 2022 anti racism strategy which focused on three pillars. One, creating an empowered internal office to challenge systemic racism. The second was building a community collaboration table to amplify grassroots voices and the third was securing sustainable funding. Alongside this, we continue to advance our blueprint for violence prevention, our migration action plan, our commitments to inclusive spaces by developing a 2 SLGT QIA space framework. Our underlining objective is to create the mechanisms that reduce all forms of hate and extremism, not simply respond to incidents, but strengthen resilience before harm escalates.

We are sending a clear message to to the whole city. In Edmonton, there is no place for hate. Next slide when we speak to our diverse communities, hate and extremism are continually raised as critical concerns requiring the most support. From a municipal perspective, our role in preventing violent extremism begins long before criminal thresholds are met. Cities are often the first to experience shifts in social cohesion, whether that shows up as increased hate incidents, rising tensions between communities, or a sense of fear that is expressed in neighborhood spaces.

Residents do not differentiate between levels of government. When people feel unsafe, they call the city. Because the landscape is becoming increasingly complex, with rising polarization and doxing of public officials, there is a growing pressure of the city to find new ways to respond. This is why we're so pleased to partner with organizations like the Strong Cities Network to innovate on how we address hate. Our interest in AI emerged from a practical prevention lens.

The question was not about surveillance. It was about whether or not we could provide understand emerging patterns early enough to activate supports before escalation. What we consistently hear from communities most impacted by discrimination Discrimination is that trust in institution matters as much as the out as the outcomes. Trust itself is the key outcome to operationalize factors we lead with trust. We we are creating a community collaboration table that is a multi sector partnership including community leaders and law enforcement and this will focus the the main focus of this will be addressing hate crimes.

We are responding to the community's desire to play a proactive role in prevention rather than only being engaged after the incident has occurred. At the heart of this approach is our core hypothesis that building social cohesion is our strongest defense. To help prevent violence and extremism as the building block supporting the foundation of cohesion, we are exploring how artificial intelligence can play a limited and governed role in countering harm in Edmonton. This is very practical. We focus on situational awareness, understanding the environment rather than targeting individuals.

We are focused on identifying trends on open source information and community level data, not individuals. The initial exploration involved a pilot project focused on collecting data on hate symbols in public spaces. The data collection tool established a shared reference library for our peace officers. The internal anti racism office complemented this very first step by funding community art murals in those frequently targeted locations, proving that a true response must address both the harm and the sense of belonging. Another tool that the city of Edmonton currently uses is called Unison.

Unison is a public safety deployment technology that provides two interrelated tools. By collecting and standardizing demand data from various multiple departments and community partners decide by Unison uses AI to predict hotspots and guide proactive resource allocation and track by Unison allows partners to Monitor neighborhood demand trends and assess the impacts of their actions. Unison is a system that consolidates non identifying safety data across the law enforcement and community partners. Built on privacy by design and data minimization, it intentionally excludes race, gender and personality personal identifiers to prevent discriminatory feedback loops. AI does not replace community engagement, it does not replace human judgment, and it certainly does not address the root causes such as systemic inequities and social marginalization.

What it does help us prioritize, it helps us move the reactive response to earlier coordinated action, engaging community partners, social agencies, youth serving agencies, and when necessary, necessary other system actors. Lastly, we explored a new concept of scanning public online information to create predictive capacity for community safety. While the idea was promising in principle, it immediately raised vital practical questions about how our mandate, legal authority, community information versus national intelligence and governance. Our initial exploration received broad support conceptually. At the same time, partners raise important governance questions.

Privacy, jurisdiction, cost scale and response capacity. Those concerns are not obstacles. They are part of a responsible public sector innovation. Ultimately, our work proves that while data can help us understand the patterns of hate, only a foundation of community trust can help us deconstruct them. For us, AI support is a supporting tool within a broader prevention ecosystem grounded in human rights, transparency and community legitimacy.

Thank you.

Think about how a city might try to conceptualize this within the kind of boundaries that you have and some of the challenges that you have in making this work in a city context. So helpful. Before I move on to our last speaker in this panel, I just want to remind everybody that if you have questions for our speaker or a panelist today, please do go on the QR code and add them in and we'll get to them at the end of the event. So now, turning to our last speaker in this panel, and we're making another shift to a completely different direction. We're going to be now looking at the from an online platform perspective.

So I'm now pleased to welcome Mr. Henry Adams, who is the Director of Trust and Safety at Resolver. So, Henry, you sit very close to how risks include including violent extremist exploitation of online spaces are identified, assessed and escalated. Could you briefly introduce how AI is being used or tested to support PCVE relevant functions at Resolver and what has strengthened PCVE outcomes there?

Sure. Very, very happy to. And first of all, thank you of course, to the UNOCT for inviting me to speak today and the Republic of Korea as conveners. It's an honor to represent Resolver here. As mentioned, my name is Henry Adams and I'm Director of Trust and Safety at Resolver.

And if you're not familiar, we're a global online safety technology intelligence and advisory partner. And perhaps unusually today I want to speak to a message of positivity and of hope in leveraging AI for good. Resolver, we support platforms, online safety regulators and other key partners with critical intelligence across what we call the full safety life cycle. Now, by life cycle I mean that we use AI and critically expert human analysts to support designing out harm, detecting it as it begins to manifest defusing risks through shaping product policy and feature interventions and supporting stronger disclosure and lesson learning to inform what is a continuous cycle in practice, but something we've been doing now for more than 20 years in delivering our mission to make the online world a safer place. Now, as I speak today we specifically use AI to prevent encounter extremism in multiple ways.

It's easiest to conceptualize this using what we call the ABCs of online safety, identifying actors, behaviors and content using carefully trained AI. Now, we often only think of the latter content, but actually all three can benefit from using AI to detect risk. And when you leverage AI across all three, you strengthen the outcomes quite significantly. So first, we use AI in our standard natural language processing enrichments and ongoing refinements. We align these to emerging extremist threats.

This is some of the oldest use of technology within Resolver, but it continues to play a critical role at the earlier stages of of our intelligence operations, filtering out noise and surfacing genuinely relevant extremist related content actors and behavior patterns for our subject matter experts to review and alert to partners for preventative action. And there's an important note here that action means a wide variety of approaches, including but well beyond just moderation and takedowns now. Second, we leverage machine learning classifiers, which we've specifically designed to identify potential glorification of bad actors and groups. And this is most critical when we're supporting partners to respond to unfolding critical incidents and attacks where calls for violent reprisals often otherwise go viral. Third, we're customizing and deploying multimodal large language models capable of detecting extremist symbols across video, image and livestream content.

Now, this approach lets us detect these specific symbols, insignia and iconography that is often used in that wider gamut of promotional and base building content. Fourth, AI is increasingly a critical component of detecting hybrid risks. We're seeing offender and victim ages sadly getting younger and younger, and the marked role of isolation, sadism and nihilistic extremism in offending as Part of a constellation of other harms and offences. And in this newest case, we're layering all of our AI classifiers together across risks and platforms. Now this is critical for some of our prevention work.

One platform may have a signal of self harm imagery, another for a different platform has text based child grooming for the same group. And another contains extremist ideation. And yet another features animal abuse content. Now, an extremist only AI classifier could barely scratch the surface of that particular case. It's not enough on its own.

We use AI to help us better across cut across these multifaceted harms and appreciate the wider risk in front of us. Because we can only act on what we can collectively see beyond our self imposed and often quite definitional silos. And one compelling example of this is the comm or 764, an issue on which we released a major report just two weeks ago. But we don't exclusively use AI for tactical detection and decision making. We leverage it also as a strategic tool using a technique called retrieval augmented generation.

And that aids our analysts in surfacing, interrogating and analyzing the longer form intelligence we've produced. So we trust it over many years, helping to identify complex new trends faster. And then finally, we do none of this work alone. And of course, all technology is ultimately human. Overall, our human experts are using AI to help detect threat actors, risky behavior and extremist content.

And we use these intelligent signals to support our partners to better design platforms and features, detect and diffuse threat, risk and harm, and importantly, better disclose what we've learned from our collective PCE efforts across the world. And the practice guide that's being launched today touches on much of what I've just said. And so I do offer my congratulations for a very excellent contribution to this space.

Behind the curtain. A little bit on how platforms are managing managing this risk. Based on your experience, what safeguards, oversight or partnership are essential to ensure AI supports prevention without reinforcing bias, stigmatization or over policing or over surveillance.

So let me start by emphasizing we should not. In fact, we cannot rely solely on technical solutions to detect all potential extremist risks. We have to layer and situate what we do in a broader set of approaches and an expert human touch. Now, if we fail to layer our efforts, we increase the likelihood of only finding repeat occurrences of issues we've seen before or very similar ones. Now that might sound like an ideal outcome to some specialists, but in practice it creates a self fulfilling detection loop in which we become more myopic, focused on a smaller subset of risks and start missing, emerging and evolving threats, especially in a world of lone threat actors, and especially where we're failing to provide for cultural and linguistic expertise, which is where Resolver specialises in particular.

Now, a key path to avoiding this is to carefully consider consider how we train and how we deploy AI. The old adage of garbage in, garbage out has never been truer. It's important to have a golden data set, specially labeled data verified, quality assured by expert humans, with the outputs continuously reviewed during and at end stage. Now, when we say human in the loop, the word in is really important. It can't just be having a human at the very end, or indeed the very beginning.

I believe we need humans through the loop and these tools should never be designed to set and forget, given how sensitive their deployment is. And then finally, sharing our knowledge between institutions is also essential moving forward. And at the same time, we have to carefully control access to AI systems with appropriate guardrails to enable legitimate analysis of extremist Related Content.

I think today's discussion again this panel just really highlighted again that, you know, AI and PCV is not a future issue, it's a current issue. It's already being employed in array of in an array of contexts. And it also showed, I think, that responsible AI in PCV is context specific and must be grounded in trust across youth spaces. Local government and trust and safety functions, safeguards and human judgment just remain really critically essential. So thank you to all of our speakers for sharing their practical insights.

So now to move to the next session, I would like to re welcome Mr. David Wells to the floor and also welcome for the first time Mr. Samuel Segan, who is both of whom were the expert consultants responsible for the UNOCT practice Guide. And before David and Sam take the floor, we would like to share a short video highlighting the AI and PCB project.

AI offers enormous potential but also serious risks. It can accelerate hate, spread misinformation and disinformation, and deepen division, fueling conditions that contribute to violent extremism conducive to terrorism. Terrorist groups such as the Daesh are already exploiting AI and digital technologies to expand their reach, groom and recruit across borders. Yet AI holds significant opportunities for peace, resilience, building and preventing and countering violent extremism conducive to terrorism. It can support PCVE early warning, improve analysis and evaluation, and help Member States deliver more effective PCVE responses in line with the United Nations Global Counterterrorism Strategy.

That is why the United Nations Office of Counterterrorism through its Global Program on Preventing and Countering Violent Extremism, with generous support from the Republic of Korea, is leading new efforts to harness the potential of AI. Through research, global dialogue and mapping, we have identified trends, challenges and emerging solutions at the intersection of AI and pcve. Our new practice guide is designed to help policymakers and practitioners use AI safely and responsibly in PCVE policies and programming based on principles of human rights, gender equality, transparency, accountability and do no harm. It provides practical guidance and recommendations to mitigate risks of AI and harness its potential future. Future training will turn guidance into action, enhancing capacities, resilience and collaboration.

This is only the beginning. Together we can shape the future of PCVE with the support of AI.

Thanks, Sian and yeah, before I start, just another plug for the QR code for questions. I think based on what we've heard today, I'm sure there are lots of questions that you all have. I'll just wait for two seconds for the slides. But yes, I'm going to talk to you about the guide and the projects as a whole and just briefly at the start at least, give you a sense of the timeline. I think the main thing I really want to highlight here is the amount of consultation that took place as part of the development of this guide.

This wasn't myself and Samuel sitting in in a remote working space just beavering away. We did really want to get insights from people working in the field, practitioners, policymakers and other experts on this. So I'm going to touch on one of those components, the survey, in my remarks. But I also wanted to mention as well the pilot trainer trainer curriculum, which we took place last week. And Samuel will be sharing some of the reflections on that in his remarks.

If I could go to the next slide, please. So most of you, or some of you have a copy of the physical copy of the guide in front of you, but this is kind of the overall structure. I particularly want to highlight the workbook number six, because we really wanted to make this as practical as possible. So there are some resources that we were testing in the Train the Trainer curriculum last week. But I really wanted to touch on the survey because I think one of the real challenges with this, and this is why we're so grateful to the Republic of Korea for the funding for this, is that as Sian said in as we've heard already, people are already using this technology in pcve, but it's very hard to grasp how widely who's using it, what they're using for.

And so if you could go to the next slide, please. We really wanted to start the project by reaching out to the community who work in PCVE to get a better understanding of exactly those questions, some of the challenges they face, some of the barriers to entry. So we launched this survey over the summer. We got a pretty good level of response rate think and a pretty regionally diverse response rate as well. Good gender balance.

The median age may be slightly older than might be expected. I think it was really interesting hearing Maximilian's remarks earlier again in terms of whether that experience might be different with a younger group. We'll go to the next slide, please.

So one of the things we really wanted to understand before we developed the guide and obviously developed the training curriculum was to get a sense of what, what people are using AI for, if they're using AI, and how they rated their self or self rated their skills and experience. I think the number was maybe a little bit lower than we were expecting. Again, we were as representative as possible, but obviously there may be people working in PCV that we didn't contact. But a relatively low rate and it was particularly interesting that the rate was significantly lower when those respondents were working in government organisations. So again, as we start thinking about barrister and entry organizational readiness, that the number for government organizations was around 10%.

Self rated skills, as you see, you know, relatively low. I think it's interesting that the use of AI tools themselves, obviously people are using them in their personal life, their private life is higher than the other two. If you go to the next slide, please. So one of the other things we did, we wanted some, not just some data based questions, but more like actual contextual data that we could use to develop the guide. And so I really wanted to highlight some of these risks and barriers.

We've heard about some of them already today. But there are lots of practical reasons why people in PCV aren't currently using AI and I think it's really important to understand those before you can design solutions to help them overcome them. So I think just to highlight the ones that are on the screen inaccuracies. I'm sure all of you who've used AI tools are aware of the hallucination issue, but I think we really wanted to highlight the lack of contextual knowledge issue, particularly because of the training data and we'll come on to some of those in Samuel's remarks. But so many of these tools are based on information openly available on the Internet and there's huge biases in that.

In terms of languages, certain cultures and languages are not really represented on the Internet at all. And so some of the PCV practitioners were telling us that when they tried to use these tools in their own context, it just wasn't effective. We heard a bit already about privacy, but certainly a lot. Again, respondents were really concerned about the sensitive nature of the interventions they're designing. Vulnerable individuals and where that data is stored, how it's shared, who has access to that.

And particularly in certain country contexts, you know, do law enforcement, do the governments have access to that data? So that was a real concern as well. The transparency and explicability we've heard a lot about today already. So I won't labor that point too much. But again, I think not just those practical barriers, but more operational ones as well.

So, you know, is AI actually going to be effective in that environment? I think for some people, they really did say that we really want to prioritize that human connection and that actual provider provision of services in a real meaningful way. And I think again, as Maximilian touched on that lack of human connection is such a problem already. So moving towards tools that maybe exacerbate that doesn't feel like a good idea. I've touched on this and again, we'll come on to that in the next session.

But this restrictive organizational policies, I don't know how it is in your own national governments or your own organizations, but for a lot of people it just felt that they weren't clear on what they could or couldn't use it for, or there were restrictions that were in place that were inhibiting their innovation. And then finally the resource question, I think AI is seen as a technology that will answer all our resource problems because it's free at the point of view, so very cheap at the point of view. But I think for a lot of people there were some technology issues, but the training in particular was an issue. And we'll come on to that in the final slide. But I think also the human in the loop that we've heard so much about today, in practical terms, the kinds of interventions that some of these practitioners are working, and you need humans in the loop in a very resource intensive way.

So you might introduce an AI tool, think you're saving lots of resources, but if you actually want to provide the oversight that you need, you actually need potentially more humans to check in on the outputs, the accuracy, context, etc. Etc. So, so there were some real risks and barriers that were identified and again, we touch on those in the guide itself. So if you just go to the final slide for my part of the presentation, so this, I think, most importantly, both in terms of driving the curriculum itself, but also thinking about next steps. A huge number of people have never received any training at all.

And if you look at the numbers thinking about AI, specifically in pcve, but also the human rights and ethical issues, the numbers are pretty high. So there's a clear need and a clear desire, I think, from the community that we reached out to, to use AI, to integrate it where possible, but just a lack of knowledge about how to take those next steps. Obviously, the practice guide we hope will help fill some of those gaps. The training we delivered last week was a success and we'll hear more about that later. But there's a clear need for a lot more in this space.

There's a real demand for that. So, yes, Republic of Rio has been a huge supporter, but I'm sure there's more work to be done and I hope other governments stay forward in that regard. Samuel, I'll hand over to you.

Thank you, David. So what I focused on was particularly the opportunities that this presents.

So in the next slide you'd see some of the use cases that we highlighted in the guide, like early warning, monitoring and evaluation, direct engagement to like maybe chatbots, to be able to point at risk individuals to valuable resources outside of just the chatbots. We looked at training and optimization, how creating scenarios might be able to help people, at least new PCV practitioners learn how to engage and use some of these tools. We also looked at the role it plays in research, particularly researching large volume of data, both text data, audio data and video data and what that would look like like we also looked at the, and I know many organizations here do this online monitoring and what that would look like and the impact it would make. We also can see that how AI tools are currently being used for like media detection, the rules like picking up deep fakes and the like, and how to create the right crisis communication to at risk individuals, as well as creating the proper like messaging campaign. In the next slide, I kind of talked about some of the risks that we already see that we highlighted.

This is not exhaustive. There's still quite a few other risks that I guess we haven't actually considered given the fact that the technology is still very much evolving. We looked at the risks to like fundamental human rights like right to privacy and what that would entail if you have to train models on specific kind of data and what that would pretend and issues like maybe content moderation and what that looks like for freedom of expression, particularly for people in low resource areas. With low resource languages where they're not very represented in the models. And what that means is that models that are trained in say a Western context would oftentimes flag some of this post and get them pulled down.

And what that would mean for the freedom of expression is also something that we took into consideration. We also looked at some discrimination that could happen in the algorithms from, not just from the data set, but the design of the algorithms itself and how that, and from the feedback we got how that could play a role in terms of context. So the context bias, what context bias would mean would be training a model somewhere and deploying it in a different context. And so we see that some of the PCB practitioners we spoke to a bit worried that the AI systems oftentimes do not have very contextual knowledge about what they're doing, the areas they live in or in the countries that they work in. And this is, I mean this is not a very unique case.

The common crawl, which was like the data set used to train much of the large language models we see today, was heavily like Western ideals and Western document segmentation and languages are heavily represented in it. Over 70% to 80% of the data set were trained in Western languages, particularly Indo European languages, English, German, French and Spanish. And just about 4% of the data, even though it's about 300 billion pages of data, were trained in other languages. So the challenge that that would present is when you try to query some of these models and to create scenarios for say PCB work in low resource areas. Oftentimes they provide very misrepresented data sets and you see things like hallucination take place.

We also look at some of the ethical concerns, the possibility of weaponization of these tools. And we also looked at the impact to have on climate. I know lots of the big tech companies are actively doing work to try to reduce the their carbon footprint by using renewable energy to power some of the data centers, create new coolants and cooling systems to be able to ensure that there's less use of fresh water to cool the data centers. We also looked at the need for human oversight and the loss of control and what that would look like as well, and the need for transparency, accountability and explainability in this models. And also there's the unintended psychological harm that could come about.

So sometimes trying to off ramp like at risk individuals could have like adverse effects or creating narratives using AI systems could create like the adverse effect from what you're expecting them to do. We also looked at, in the next slide, the risks and some of the mitigation strategies. We talked about the need for like privacy, so privacy by design and even things like analyzation of data or using pseudo pseudo names for sensitive data. We looked at what that would look like for security, what that would look like for freedom of expression. We also talked about, I know there's been a lot of talk about human in the loop where we can introduce human oversight and control.

We also looked at AI risks and impact assessments. So in the workbook we have a risk assessment that there to guide PCB practitioners on how best to identify possible risks that would happen when they use AI systems. We also look like the need for governance, auditing and accountability. Talked about multi stakeholder approach to engaging. So in the workbook we also have like a stakeholder mapping exercise that would help PCB practitioners to be able to identify who those stakeholders are and what that what the AI system might have an impact on them or their work.

We also looked at some proactive and adaptive responses to AI tools as well as some responsible AI practice that we continue to see like red teaming, like adversarial testing and all that. That will be very essential. Next slide please. We looked at what implementation would be as a next phase, things like building organizational readiness. We have an exercise in the workbook that is focused on organizational readiness assessment where we assess certain skill level, assess existing technologies that civic organizations might have and what that would look like.

We also look at some essential competencies that would require some literacy, some AI literacy levels like you could see in the, in the results from our survey. Not too many persons are actively using this. And then we ask for what would that look like? What sort of skills would they need to learn on the job to be able to improve the output of their work with pcv? And we also looked at some of the implementation frameworks or some strategic planning, some readiness and needs assessment.

Is it essential to use AI tools? I think that's one of the questions that many times we don't ask ourselves like do we really need to use AI here? Or we could just do it, have some human intelligence interaction and probably will produce better results as well. We looked at the need for like piloting interventions and not just going all full out there using AI tools as well. And lastly, I know I've mentioned some of these but we've got, in the next slide you'll see what we've got.

Like in our workbook we've got the AI PCB Resource guide. So for practitioners they can have access to some of the guides and we added some links that might be very useful for them as well. We looked at the organizational readiness assessment, risk assessment, we looked at the human Rights and ethical guideline checklist. So basically just walk you through the process of looking at some of the things that would be essential to consider before the deployment of AI tools. And lastly, in the next slide, you'd see our pilot training, which we did last week.

We got some very useful feedback and comments from some of the participants. And particularly I think what, what's important to highlight is the affirming, the need for very practical way of engaging with identifying risks, practical ways of identifying stakeholders and activities that will be required. And we'll be able to do that with the workbook. And you can see, see some of the comments on your right there. Like, the exercise gave us a better understanding of how to evaluate gaps in our approaches and to be able to reevaluate the methodologies when it comes to using pcb.

So we are glad to the Republic of Korea for their support in helping us get this far and the feedback that we got that has been able to help us produce this guide. And so thank you very much.

Thank you so much. I'd now like to pass the floor to my colleague from the Republic of Korea, Mr. Yoo Chenyo.

Thank you. So as we move into this final segment, we want to shift the focus from whether AI can be used in PCVE to a more foundational question. Are institutional actually ready to use it responsibly? Throughout today's dialogue, we've heard about the opportunities AI presents for prevention, but also the risks when governance, safeguards, skills and human oversight are not in place. The Republic of Korea and UNOCT have taken this insight seriously.

And this closing session also marks the announcement of the next iteration of our joint work, which will focus squarely on organizational readiness. This segment therefore looks ahead at leadership, institutional capacity, and the practical conditions that must be in place before AI is meaningfully integrated into PCB policy or debate. I'm pleased to be joined by Dr. Bailey Parnell, founder and CEO of Skills Camp, and Ms. Lily Vincent, program management. Officer with the UN Office of Counterterrorism. So, Bailey, you work closely, closely with organizations navigating AI adoption in high stakes public interest environments.

From your experience, what do we often misunderstand about what it really means to. Be ready for AI? How do you see this applying prevention context?

Thank you for having me. I'm glad to be here. As a Canadian, but living in New York and through Skills Camp, we work with partners really all over the world. So when we talk about AI readiness, I find that many of us are often jumping to data or policy or platforms. But what I have found in every organization that we've ever worked with, readiness never comes down to the platform.

It comes down to the people, and more specifically, the people leaders. And I think that we're at this very unique time of history where many of the PCVE organizations and really organizations cross industry, are adopting AI for the very first time. And so we have this unique opportunity to help them do it right before habits are entrenched, because it's a lot harder to fix things once it's already gone wrong. And so, by focusing on organizational readiness, I think that the UN Office of Counterterrorism and the Republic of Korea are actually quite ahead of the curve from what I've seen, because they're asking a question that is often missed, which is, before we deploy this magical new technology called AI, are we legally, ethically and human humanely ready to lead it? And in practice, There are three pillars of readiness that I work with.

The first is governance. Are we legally and ethically ready to, quote, unquote, turn it on in their regulatory environment? The second is technical readiness. And this is, do we even have the digital literacy to know what to do with it once we do turn it on? And the third, the most forgotten and where skills camp actually started as an organization is human readiness.

And are people afraid? Are they afraid of doing it wrong? Are they afraid of it taking their job? Do they have critical thinking skills? Do they, Are they afraid of change?

Are they tired of change? Are they asking themselves, am I making the world a worse place by doing this? And it cannot be understated because if you're missing one of these three pillars of readiness, it will break down as a system. If you're missing governance, people will use AI anyways and what we call shadow adoption. And they.

I have actually had quite literal research participants tell me, I email my stuff. I email myself stuff at home. I use public AI and I email it back to myself because I'm not allowed to use it at work. If you're missing technical readiness, you'll be like a healthcare consultancy we worked with recently in a bank. And they turned it on.

They invested all this time and money, and then a month later, no one was using it, having wasted all that time and money. And if you're missing the human readiness, then a quiet dissension, mistrust, sometimes going into fear and disengagement grows, and it spreads faster than any of the technologies. And so this is why we would always start with the readiness assessment in order to make any intervention useful. Another practical example would be a pretty famous mental health hospital in across Canada research hospital. And we were, I mean, I was kind of surprised to find that they were actually way further ahead on the governance pillar than even some of the tech companies who you think would be maybe further along.

But they had a very human motivation. Their motivation was we have a ton of clinicians, we actually want them to have more time to spend with people in relational work and in the community. Can AI help with this? And this was the motivation. So their intervention was wholly more focused on technical readiness and human readiness because not all these clinicians actually work with AI or know even screen work all the time.

And sometimes the humans were asking questions like, oh my gosh, is this just going to be another thing on my plate that I have to deal with? And then that was withholding their adoption. Or it was, you know, is this going to take my job? Like, am I supporting the end of my own career? Was the human readiness pillar.

Right. So and so it didn't matter if they turned it on and did all this, you know, regulatory work ahead of time. And above all of this, and at this institution sits the leadership tier. And it doesn't matter if you're an NGO or an organization or a member state. When humans come together, it is the leaders in those groups that decide that interpret this unknown, that decide what skills that we need to get there, that either model a culture of exposure, experimentation, or don't and really decide if AI is going to be something that they are going to fear, going to ignore, or going to use for good in that organization.

And so why would we focus on leaders in the next phase? And why do we focus on leaders quite a lot is that they are a high influence group. They're the connective tissue of readiness. And if you train one leader really well, you will cross all three pillars at the same time in the organization. And so when the leaders know what they're doing and the organizations are actually ready, then PCVE quite literally moves faster and more effectively.

And this is the stuff that is, this is why we're here really, because that's the really motivating part of what we're talking about here is, is a leader who might use AI to summarize field notes in seconds, still, still with their eyes on it so that they can get back to community and relationship work, or the team that is using translation technology to reach a youth in their unique language where they're at when they need it, still keeping the human throughout the loop, or the department that is analyzing a ton of data, kind of identifying where misinformation trends are happening again, keeping their judgment to make sure that it is what it is supposed to be. And I think this is what readiness looks like. It is competence and confidence with conscience in an organization. It is competence and confidence with conscience. And really, now is the time for us to do this.

We're at a very strange time of history. You get the privilege of being in this room guiding where this might go and how we might be talking about it 50 years from now. And so if we can get these organizations ready and their leaders ready, if we can help them get those governance frameworks and the technical literature, literacy and the human capacity, then AI will not replace the human heart of pcve. It will actually amplify it. And that is absolutely incredible.

And especially in these times when we know that the practitioners are facing budget cuts and it just feels so tense and stressful all the time, we actually have a really incredible opportunity to help them get more time to listen and to learn and to. To lead these prevention efforts with grounded in dignity and human rights. But this only happens, of course, if we teach good people to use AI for good. So I'm proud to be a part of it as well, and I thank you for the opportunity to make this happen. Thank you, Bailey.

What you've outlined, governance gaps, technical literacy challenges, and human concerns around fear interest closely mirrors what we heard from PCV practitioners during the first phase of this project. So here's where I turn to Lily. So what is your take on how these same readiness gaps emerge throughout UNFCT's work with PCVE practitioners?

Thank you, Yu Chun. And thank you, Bailey.

So it's quite lovely to hear you say that, Bailey, because all of the three points that you've highlighted, the three pillars really ties very closely to what we found found in the process of developing the practice guide on AI and PCVE and what we've observed in, in our interactions and consultations with the government sector, but also practitioners working on pcve. As we learned from the previous panel, one of the clearest findings from the global survey underpinning the practice guide is that AI adoption in PCVE remains limited. And it's not necessarily because of a lack of interest. As we learned From Dave, only 23% of the respondents we heard from reported using AI in their PCVE work. And among the government actors that were part of that survey, it was actually even lower than that percentage and when we asked why, practitioners pointed to three main constraints, again linking directly back to what you've highlighted, Bailey so the first was around organizational readiness gaps.

So this included the absence of policies, government frameworks and clarity around accountability for AI enabled decisions. The second was capacity gaps, so limited AI literacy and lack of confidence in assessing risks related to bias, data protection or human rights impacts. And the third is around human hesitation, concerns about trust, legitimacy, reputational risk and the fear that doing harm in sensitive protection context. So for our space, I think in in particular this, this really stood out. While distinct, these barriers reinforce one another and organizational readiness sits at the core.

Without governance, clarity and leadership oversight, building skills and confidence alone will not translate into responsible adoption of AI in the PCB context. And we are also seeing more broadly in AI the AI PCV space that risks are evolving faster than institutional responses. So as we've heard today, AI is already lowering barriers to terrorist propaganda production, enabling rapid narrative experimentation and amplifying disinformation. And at the same time, PCVE organizations are under pressure from resource constraints and rapid technological change to keep up. This combination creates a range real danger either in the premature adoption without safeguards or complete disengagement due to fear and uncertainty.

And both outcomes undermine prevention objectives. And this is why organizational readiness has emerged as a critical missing layer throughout this project and why the next iteration of our work focuses squarely on strengthening the organization's ability to decide if, when and how AI should be used in PCB contexts. And in concrete terms, this means supporting organizations to conduct structural AI readiness assessments, engage in their leadership to make informed decision making, clarify governance arrangements and embed human rights and do no harm considerations into organizational processes, ideally before any AI system is deployed. It also means normalizing the idea that in some contexts the responsible decision is is not to use AI at all. As Samuel had pointed out earlier.

I might stop there. Yu Chan, back to you.

So building on what Lily has just. Outlined from the practice guide, may I. Ask to Bailey, what is one concrete readiness issue that PCVE leaders should address first and what tends to go wrong.

When that issue is overlooked?

Bit of a bias here because I started in this consultancy, started in teaching people skills like soft skills like communications, and realizing that it was the thing that made success in any forum, but it wasn't taught in school. And so I think the thing most forgotten that all PCVE organizations, and really all of our organizations too should be thinking about is the human readiness. Because it doesn't really matter what the change is, if how the change is Communicated from leadership. Even theories of what I would do in my world is like theories of transformation and how are people feeling like they're going to be supported? How is it communicated the first time?

When is it communicated? Is it alongside other stressors, actually matters a great deal to whether or not the humans are even going to adopt it or think that this is the worst thing ever. And so that part is forgotten. The part that's also forgotten is in that human readiness space is how we experience change. If there are.

If we had this most amazing plan to roll out AI and we've done the governance and the technical readiness we have, we're so excited this is going to amplify PCVE efforts and then something happens in their area that they're dealing with and everyone is coming to work stressed, your brain is quite literally compromised. You don't become a different person because you entered the virtual office that day. And so if you're compromised on, on like a, on a physiological level, there's no way that you're making the most informed, most creative, most humane decisions of your life. And so I think that the human piece of it in AI space, because AI was often coming out of the tech, out of the tech world, is the part forgotten and most important in the. In the right humans using this for the right reasons.

Thank you. I think we are approaching to the. End of the conclusion. But Lily, from a practitioner's perspective, how. Is UNOCT translating these concerns, organization readiness concerns, into concrete support for PCB institutions going forward?

Thank you, Yuchan. So maybe I'll just close very briefly with a practical reflection on the experience that we've had over the course of the project and then an invitation to everyone in the room and online. So as we know, one of the key lessons from the Practice Guide is that responsible AI use in PCV is not solely a technical challenge, it's an institutional one. And it requires leadership, engagement, internal coordination, and the willingness to ask difficult questions about risk, accountability and legitimacy. So the next iteration of this work will focus on helping PCVE organizations operationalize the guidance already set out in the Practice Guide.

There is a readiness and risk assessment tool in the workbook that could be a starting point. And we're looking at supporting organizations to adapt them to real context, so working hand in hand with them. And again, this works with working very closely with leadership teams to support governance development and creating spaces for people learning across institutions facing similar challenges. So for those in the room and those online, across member states, implementing organizations and partners, this is an opportunity to help Shape emerging practice. In the coming months, UNOCT will be engaging with interested organizations to contribute to this work, to share lessons learned and participate in structured policy dialogue on responsible AI governance in pcve.

And we know that AI will continue to involve whether institutions are ready or not. And our collective task is to ensure that PCV actors are equipped to engage with it deliberately, cautiously and in line with human rights obligations, rather than reactively or by default. Thank you.

Thank you, Lilly and Bailey. So, just to wrap up this session, this discussion, I believe, has reinforced a.

Central message that we would love to deliver to today. Responsible AI use in PCB begins not only with technology, but with institutions that are prepared to govern it, understand it and apply it with care. On behalf of the Republic of Korea. We are pleased to continue this collaboration with UNCT and to support the next iteration of this work translating global principles into practical institutional capacity for PCB actors. Thank you.

And I would like to pass, like to share.

Left. We actually have got a significant amount of questions. We won't be able to get to all of them from the QR code, but we will try to get to around 3. So the first question is from the Russian mission to the United Nations. Thank you, Oleg.

And the question is to the panel, but I will direct it. Was the private sector involved in drafting this guide? And how do you see technology companies being encouraged by or required to implement its recommendations? In practical terms, how can we ensure that AI tools are not misused for terrorist or extremist purposes? I might start with Dave on how private sector was involved.

Thanks, Sian. And thanks to the Russian representative for the question. I think in terms of the consultation exercises and the peer review feedback for the practice guide, we did have representatives from the private sector, but I do think this was something that we want. We wanted more of, I think, particularly for, and I'm sure Nagam can speak to this, but particularly for private, for AI companies themselves. They're not necessarily currently active in the PCV space or thinking about terrorism in the way that big social media companies have been for a really long time.

And so I think we're really keen to have more opportunities to bring them to the table and have that dialogue. It's certainly the guide is designed primarily for PCP practitioners, but there are recommendations in there for the private sector too. And so I think if that engagement could be fostered through a unocc, that would be a really positive thing.

Definitely engaged, if anything, honestly, I think that the work that's been done since kind of moderation has been ramping up within the context of AI has has picked up speed in a way that, you know, all of your colleagues here can also probably attest to it throughout our working groups seeing, seeing that that kind of up ramp. So I would say this is complementary. And kind of a continued conversation and part of the bigger process, but definitely a super important piece for sure.

I think the only thing to add. Is from throughout this process for us the the starting point was really to hear from PCV practitioners and policymakers. So that's really the framework that we came from when we were developing the guide, especially when we were starting out. The survey was really instrumental in helping shape that and I think some of you in the room and online as well really took part in that. And through the process, as Dave mentioned, we did consult with tech, but that really will be more of an engagement that we'll continue with in the next phase of this project.

The next question I'll go to is from the permanent mission of Sao Tome and Kunshipe and the question is with the human flaw of the individual often engaging in confirmation bias, we often see individuals engage deep fakes because it validates their pre existing beliefs while dismissing any other contradictory source of information as being false. Even if we do manage to combat deepfakes and AI, how do we address the human mentality of people who do want to believe in the deepfakes and radical messages being sent? It's a great question. I'm going to go to Matt first and then Maximilian if you'd like to follow up. Yeah, thanks for that question.

I think that the challenge here is that the nature of how people are desiring to consume and engage with content and technologies is shifting. And so in a lot of respects we have to just adopt a baseline acceptance that we may not be able to shift them out of that uptake. Instead we might need to think about steering or guiding that. I think. You know Max, you spoke about the idea that we can provide off ramping or diversions for individuals in these spaces and that that's probably the most viable option we have at this time is thinking about diverting them into a more pro social or positive endor engagement with that content and type of technology.

The other is engaging in what some prevention workers call pre bunking. So I think, I think thinking about that, the sort of understanding that these people need to be taught in some respects that they're being manipulated by the technology and that exposure of that reality might be helpful to fostering a better relationship with it overall.

Echo the whole point about media literacy and really fostering a sense of critical thinking. But I think that deep fakes and artificial intelligence and AI generated content often exploits human vulnerability, which is really why we need to develop those skills. And I think one solution that we were thinking about throughout my work was labeling AI videos and content. But one of the problems with that is what happens when a video slips through the filters and then you don't have that label. But people have developed a dependency to seeing that label next to AI generated content.

And so it creates this sense of dependency on the label. And so I really would welcome any solutions to that. Henry, any solutions that you might want. To.

The peasy not so lemon squeezy. No, I mean there's a two part I do want to reiterate Matt's point that a lot of what we see in some of the these more nebulous I wouldn't use group or community, but more nebulous networks of loosely connected individuals. Almost everyone participating understands that a particular piece of content is synthetic and it's immaterial to the purpose of digesting it, sharing it, using it. It might as well be authentic. It serves precisely the same purpose.

That's all. I'd.

Going to take a question from UNOCT because it is on gender bias and we haven't had this discussion yet and I think it's a great question and I'd like to go to Sandra and also Cecilia on this. So emerging concerns with the misuse of AI by violent extremist actors pose specific gendered risks such as technology facilitated gender based violence, algorithmic amplification of harmful gendered narratives and disinformation, and the weaponization of deepfakes to target women leaders and activists. Additionally, research suggests that there are important gender related considerations for the application of AI in PCVE and similar peace and security context, such as the potential for gender and racial bias in data sets used to train AI for military targeting and other uses. Would the panelists kindly speak about some of these gendered risks and how the use of AI for PCVE can be more gender responsive? Another simple question.

Over to you, Cecilia.

Thank you Sienna, and thank you for the question. Just to start us off on that issue and thank you for bringing it up again. I think it poses kind of like a two way approach to it because we are very much aware of the fact that if we do not integrate gender from the very beginning in terms of understanding where the terrorist threat is moving, and in particular in this case, the answer would be the vernacular one, let's look at how AI it's being weaponized in the context of creating or reaffirming gender biases. And in that context, perhaps for us to be much more attuned to the risk and to assess the threat as it being posed.

I would call for an additional lens in which diverse voices and diverse individuals within the community are invited to support threat analysis. That it's not limited to government officials and that there is for instance, women led organizations being made public part so that they can voice their concerns in terms of the use of AI by state actors for PCVE purpose or other practitioners. I think the question relates back to the issue of the data set used to trained models. And I explained in my remarks the importance of having transparency in terms of how the actual algorithm that is being used operates, how it is being trained. And these issues would only continue to remain in terms of us being very much aware from the very beginning and very much transparent in terms of the type of technology we're using and whether that technology has inherent gender risks as well as other risks.

And when we develop mitigatory measures, we need to continue to be very meticulous in following up in terms of whether they have worked or not. It's not enough to create some sort of mitigatory measures from the very beginning. We need to know what the impact has been. And I would argue in this context, looking very intentionally on whether there had been the deepening of gendered biases in the outputs. Coming up for AI in the PCV context is an area that in the operationalization of these practice guys, you would be in a position to advance more data and more knowledge.

Thank you.

With your projects, Sandra, and the conception and maybe in the application.

The concepts and exploring the, the space of AI, we actually didn't have these identifiers. So as I mentioned in the presentation, we looked more of a environmental perspective. We wanted to get the lay of the land. So we, our AI team, our data science team actually removed that. That being said, we are at the city of Edmond taking an approach of more of a community trust and community building space with our strategy.

So when we are building our frameworks, when we are building our plans like the two SLGBTQ safe spaces, we look at these, these, these themes. We also have a gender based analysis tool that all city of Edmonton employees need to take and we've actually made it mandatory that it's implemented in any council report that is presented to, to the mayor and council. Is that any project, any analysis, any framework needs to have that gender based analysis embedded in everything that, that we do. So right now the city of Edmonton is looking more at the community safety, looking at well being, looking at frameworks, building that trust with community and listening to community. I will say that there is some, definitely some communities I want to mention the, the underrepresentation of indigenous anti hate that is that we are concerned that AI wouldn't pick up because these are grossly underreported.

And then we are very concern, concerned about the perpetuation of biases and stereotypes without having that human oversight. Thank you so much. Next question is going to be for Bailey and Lily, Dave and Sam. If anyone, either of you wants to answer. It's a question from the EU delegation to the UN So the question is what concrete benchmarks or indicators should organizations use to assess whether they are truly ready to deploy AI responsibly in pcve?

Very good question. There are a couple of them. I think I highlighted about two of them in my talk. One is to understand the need first, so there has to be like a needs assessment. And then the other is to sort of do an analysis on the existing infrastructure.

So it could be human resources. Do we have the availability of those resources in place? Do we also have the. The right set of skills like education that we need to train PCB practitioners that work. They're also trying to figure out how it fits into the existing technology.

So like those who work in business transformation will tell you, like, if you're introducing a new technology, you want it to fit in seamlessly into like workflows. Right. So the question is, does this make sense, like for us to include it into what we currently do and do we have the, say, the technological infrastructure to be able to include it? So I guess that those, those are the two primary things I'd highlight.

All right. I think that you have covered a wonderful piece as well as in the practice guide. There is some very practical tools that you help you figure out the benchmarks that you need to hit. However, then we get to practice. I did talk, talk about how a readiness assessment is really important so that you're not diverting resources, human and otherwise, to things that you may already be ahead on or may not be ahead on.

But then we get into what is really a transdisciplinary conversation. And when we actually decide, okay, we feel that we need to build the technical literacy of our people because they actually just won't know how to prompt accordingly to get the outcome that they we really want for the whole goal of the institution, then we're getting into what is really like Adult learning design. And then you're going to get like, what you should not do is, you know, speak at them for three hours in a lecture format trying to teach them technology and human capacity skills and you will still not get the outcome that you want. And so benchmarking there goes into like, do you even have a learning infrastructure once you've identified the skills that they need? And do you even have any, any organizational, what we call almost like the leadership operating system for the human capacity?

For instance, did you identify that critical thinking was a skill that you would even need for your people? Did you know how to assess for this in an interview, how to ask behavioral questions like, these are. So your benchmark for the question is, is this in place before we even come in? I can tell you that one of the largest tech companies in the entire world partnered with Skills Camp and did not even have leadership development or a learning org that was even comparable to some of the smallest organizations that we've worked with. So start there and then no matter what you find, you'll be ready for what we'll help you implement.

I'm so sorry. We, we actually had so much interest in this. We've got so many questions we won't have time to get to today. But I think that really indicates the really significant interest in this topic and, and I hope that this has been of interest to everyone. But as we come to the close of the special dialogue, allow me to briefly reflect on what we've heard today.

Our discussions have really underscored that AI is no longer a peripheral or future issue for the PCV sector. It's shaping the environments in which prevention takes place and it's influencing how institutions, communities and young people interact within these spaces. So throughout this afternoon, we've heard about the evolving landscape, considered the human rights and operational implications, and learned about concrete practical experiences from practitioners, policy makers, platforms and youth representatives. A clear message has emerged. Innovation alone is not the answer.

What matters is how it is governed, how it is understood and how it is applied. The launch of UNOCT's practice guide on AI and PCVE makes an important step in translating global discussion into practical support. But today's exchange also makes clear that this is only the beginning. Continued collaboration, careful institutional preparation and cross sector dialogue will be essential as this field evolves. On behalf of the United Nations Office, Office of Counterterrorism, I would like to again express sincere appreciation to our wonderful partners at the Permanent Mission of the Republic of Korea for your long standing partnership on this and leadership, and to all of our speakers and participants for contributing to such a thoughtful and substantive exchange.

We look forward to continuing this work together before you leave today. We would be very grateful to hear your feedback on the event today via the QR code on the screen. We will then take this forward for future future iterations of this work and future dialogues just like this. So thank you and this concludes our event.