Global Centre for the Responsibility to Protect

The most important potential outcome is that the UN clearly and comprehensively articulates the scope and purpose of the Dialogue, including how it will implement and operationalize the commitments and decisions established therewithin, particularly with regard to human rights, transparency and accountability. The Dialogue should serve as the public venue to transparently communicate the full scope and mandate of the Scientific Panel on AI, as well as its working methods. This should include clarity on the scope and level of its scientific guidance, including whether it will engage directly with national or regional policies and frameworks, alongside private sector and civil society standards and procedures. It should also clarify what type of recommendations the Panel is empowered to issue, which actors those recommendations will target, and how they will promote interoperability and cross-functional policy sharing with multilateral, regional and national mechanisms, including accountability institutions, standards bodies and private entities. Importantly, the Dialogue should provide a platform to discuss how the early warning capacity of the AI Scientific Panel functions, in practice, both in terms of how this body collects and monitors information, but also expectations by relevant stakeholders to translate early warning to action. The Dialogue must also institutionalize meaningful participation of diverse stakeholders, including civil society actors, particularly those working on human rights protection and atrocity prevention, ensuring their sustained and structured engagement. The Dialogue must also strongly affirm that AI governance should be grounded in international human rights and humanitarian law, while also recognizing the potential emergence of new legal norms shaped by AI-specific risks. Finally, participants should agree on a forward-looking roadmap for multilateral coordination that bridges normative development and technical governance, ensuring scalable, coherent and prevention-oriented global AI governance across levels.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

Conversations surrounding AI development and governance must account for the ways in which AI may contribute to the perpetration or prevention of mass atrocity crimes. AI systems designed, deployed and governed without consideration of social factors, can deepen existing inequalities, reinforce exclusion and marginalization along ethnic, religious, linguistic or socioeconomic lines. AI algorithmic bias can amplify the spread of disinformation and conspiracy theories targeting particular groups, helping to reproduce bias, eroding trust and social cohesion. Systems trained primarily in dominant languages or deployed without adequate local-language capacity can further entrench these risks. Some governments are embracing expansive, centralized state authority over digital infrastructure and data flows, normalizing practices of censorship and digital surveillance as a tool for repression. Authorities have used surveillance systems to track individuals, assign them value weights based on their behaviors, affiliations or beliefs and forcibly assimilate entire populations. In these contexts, the deployment of AI, often under the guise of national security or counterterrorism, raises serious concerns about their potential to facilitate identity-based persecution. Fragmented or inconsistent governance frameworks create regulatory gaps that can be exploited by actors seeking to misuse AI. Given the cross-border nature of AI systems, data flows and digital platforms, isolated national approaches insufficiently address risks that transcend jurisdictions, especially as technology rapidly develops in states with looser guardrails. Greater interoperability across national, regional and multilateral frameworks can enable information-sharing, support coordinated enforcement and mobilize political will. Governance frameworks must provide visibility into how AI systems are developed, trained and deployed, who is responsible for their outcomes and how to prevent misuse and mitigate harm. This includes establishing red lines for uses incompatible with human rights, democracy and the rule of law, particularly in systems likely to produce unacceptable harms, and ensuring AI operationalization in line with international human rights standards.

We believe atrocity prevention is insufficiently reflected in the thematic areas identified above. While closely linked to the protection and promotion of human rights, atrocity prevention is a distinct cross-cutting policy agenda focused on identifying risk factors, preventing escalation, protecting vulnerable populations and ensuring accountability for genocide, war crimes, crimes against humanity and ethnic cleansing. Systematic or widespread human rights violations often constitute key early warning indicators of atrocity risk. AI can both mitigate and magnify atrocity risks, directly and indirectly shaping how actors may perpetrate or prevent mass atrocity crimes. Used responsibly, AI can serve as a valuable analytical tool for prevention efforts. It can help researchers, policymakers and civil society analyze large datasets, identify patterns of violence, detect emerging risk factors, monitor hate speech trends and potentially improve the timeliness of responses to imminent threats. AI has already been explored to strengthen documentation of conflict-related sexual violence, assess community needs in crisis settings and support UN peace operations through data analytics. At the same time, AI can be exploited in ways that heighten atrocity risks. Digital technologies are already used to accelerate disinformation, deepen information silos, automate censorship, enable intrusive surveillance and target or dehumanize marginalized communities, particularly ethnic or religious minorities. AI-enabled systems may also be used to suppress civil society, restrict dissent or facilitate discriminatory profiling under the guise of security or counterterrorism. These realities underscore the need for AI governance that is responsive to atrocity risks and integrate atrocity prevention strategies from the outset. As such, atrocity prevention should be recognized as a core cross-cutting consideration in thematic priorities, ensuring governance frameworks are prevention-oriented, rights-respecting and responsive to emerging threats.

Governments and private technology providers are making decisions with profound political and human rights consequences, often addressing various dimensions of AI use—such as cybersecurity and surveillance, weapons systems and agentic integration—separately, rather than in an integrated manner. These fragmented approaches not only limit capacity to address AI-related human rights risks in a standardized way, but also raise questions about the willingness and ability of governments to curb their own misuse of digital technologies, reinforcing the need for national, regional and multilateral congruence around basic principles and red lines surrounding AI. Where governance structures do exist, they are often founded on a risk-based approach that sets expectations for systems deemed high-risk. However, generalized risk-based classifications are insufficient to prevent atrocity risks from emerging; context-specific assessments are needed to understand the nature and scale of potential harm. Moreover, these models frequently rely on self-assessments by providers, namely developers and deployers, to determine risk, creating a significant potential accountability gap. Embedding accountability therefore requires not only evaluating technical risk, but ensuring that human judgment, and ethical and legal scrutiny are maintained.

The AI Dialogue can play a convening and coordinating role in advancing international cooperation on AI governance by bringing together a broad range of stakeholders who are currently working in parallel but often in fragmented ways. Independent of the strength or maturity of AI-specific regulation, significant efforts are already underway across legal, policy and technical communities to address harms and strengthen accountability for abuses involving AI technologies. In the atrocity prevention context, in the absence of comprehensive AI governance frameworks, practitioners and scholars have drawn on existing regimes related to dangerous speech, misinformation and disinformation, child protection and safety, sexual and gender-based violence, anti-discrimination, and broader human rights obligations. Bringing these communities together can strengthen the breadth of AI governance and pave the way for comprehensive multilateral structures around AI. Moreover, leveraging expertise and agreed upon language/processes/policy structures from those working in fields with existing governance structures may bolster the capacity for international cooperation on AI.

The Dialogue should engage with and provide update on existing multilateral frameworks such as the UNESCO Recommendation on the Ethics of Artificial Intelligence, and support the work it is leading across its civil society and academic networks, such as on identifying AI red lines. The Dialogue should be sensitive to frameworks such as the UN Guiding Principles on Business and Human Rights, and the Scientific Panel on AI should consider cross-collaboration with the UN Working Group on Business and Human Rights to ensure AI best practices are aligned across the political and business sectors.

Different stakeholders can make distinct and complementary contributions to the AI Dialogue, and the format should be designed to capture this expertise through sustained, practical engagement rather than primarily high-level statements. Participation should also extend beyond ECOSOC-accredited civil society organizations to include independent experts, affected communities, smaller civil society organizations and technical practitioners who may otherwise be excluded. Member states and regional organizations can contribute regulatory experience, national policy frameworks, transparency measures and political commitments. It is critical that member state representation is not limited to high-level representatives, but inclusive of a range of actors from across a government's system, including political, technical, economical and judicial agents. Civil society can contribute essential expertise on the social, economic, cultural and human rights implications of AI, including how systems affect marginalized communities, civic space, labor rights, privacy and discrimination. Civil society actors are also often best placed to identify harms emerging at the community level. Academia and technical experts can provide knowledge on the design, capabilities and limitations of AI systems, helping ensure that governance frameworks are technically informed, realistic and adaptable. Their engagement is particularly important for translating complex technical issues into actionable policy options. Structured dialogue can also help bridge disciplinary divides between technical experts and political actors. With regard to format, the Dialogue should prioritize thematic working groups, multi-stakeholder roundtables and informal expert exchanges over predominantly high-level plenary sessions. Smaller interactive sessions are more likely to generate substantive recommendations and candid discussion. The process should support continuous engagement across the year rather than one-off events, with clear opportunities for written submissions, hybrid participation and follow-up consultations.

There is a persistent gap in cross-cutting expertise, particularly regarding the human rights implications of AI. Ethical expertise is important, but ethicists are not a substitute for human rights practitioners, humanitarian actors or specialists in atrocity prevention, privacy, labor rights and child protection. AI governance discussions often separate technical, commercial and rights-based expertise when these perspectives need to be integrated.

Meaningful multistakeholder engagement requires designing a program that enables the most critical expert voices to participate in all relevant discussions. AI governance depends on input from technical experts, policymakers, civil society, industry and human rights actors whose expertise often spans multiple themes. The thematic breakout discussions should therefore avoid parallel sessions on overlapping issues that force participants to choose where to engage. Both technical experts and human rights practitioners can contribute meaningfully across thematic priorities. Running sessions simultaneously risks fragmenting expertise and weakening outcomes. Scheduling is therefore a substantive inclusion issue, not merely a logistical one.

Ahead of its convening on Artificial Intelligence, Human Rights, and Atrocity Prevention, held 12 March 2026, the Cardozo Law Institute in Holocaust and Human Rights (CLIHHR) prepared a list of various frameworks, instruments and documents related to the governance and regulation of AI at the multilateral, regional and national level. This can be found here: https://cardozo.yu.edu/convening-ai-human-rights-atrocity-prevention. Similarly, ITU's Summit for Good has an AI Standards Exchange Database comprised of over 870 AI standards and related technical publications. It is available for reference here: https://aiforgood.itu.int/ai-standards-exchange/.