United Nations High Commissioner for Refugees (UNHCR)

(1) A clear and concise statement of a people-centred and rights-based approach to AI that: - Emphasises responsibility across the lifecycle - Is firmly anchored in the Principles for Ethical Use of AI by the UN - Recognises the diverse values we pursue as a collective humanity (not just efficiency). (2) The definition of a mechanism to support the translation of such a statement into governance frameworks at the national, regional and global levels.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

As a humanitarian agency, we have selected the themes we consider central to ensuring that AI governance responds to the realities of humanitarian and displacement contexts. In humanitarian contexts, AI offers potential benefits but also significant risks that must be identified, understood, and mitigated. Decisions about AI deployment should be evidence-based and distinguish between proven and speculative applications (ALNAP 2025). They should also be aligned with principles of safety, transparency, accountability, and adhere to applicable legal and ethical frameworks, including personal data protection and data responsibility, particularly given the rapid evolution of the technology and lagging safeguards. As open-source models are promoted for national DPI and related use cases, there are inherent risks, including their adaptation for purposes that undermine data protection, privacy, or other human rights, creating a governance gap that must be addressed as efforts continue to bridge the AI divide within and between countries. More broadly, forcibly displaced and stateless people are often in precarious situations where human rights protections are weakened and access to life saving assistance can be limited or fragile. While responsibly governed AI can support access to rights, protection, and services, weak oversight or inappropriate use risks reinforcing discrimination, exclusion, or harm. These risks are shaped not only by technical design choices, but also by the broader social, economic, cultural, linguistic, and infrastructural environments in which AI systems are introduced. Finally, AI increasingly drives the production, amplification, and targeting of information about forcibly displaced people, including hate speech, misinformation and disinformation. These risks can exacerbate xenophobia, violence, persecution, and forced displacement, undermine social cohesion and asylum policies, and negatively affect humanitarian operations and adherence to international refugee, humanitarian and human rights law. Prioritizing these themes is therefore essential to ensure AI governance is human centred, rights based, and responsive to those most at risk of harm.

We believe the following issues merit attention. We note that they may be included in some of the broader thematic categories included above, but would like to advocate for ensuring they are addressed within these: - Personal data protection and privacy; data responsibility; data sensitivity (sensitive data) and/or sensitive use cases. - AI and human behaviour: in particular addressing the emerging trend of para-social relationships, the high risks and knowledge gaps about how AI models may negatively influence human behaviour and decision-making, including amongst forcibly displaced people. - The environmental impacts of AI

Challenges While many challenges/risks are shared across sectors, their impact is heightened in humanitarian contexts where critical decisions and fundamental rights are at stake. Challenges include data gaps or poor-quality data leading to inaccuracies and misinterpretations; bias and discrimination; hallucinations and low reliability; personal data protection and privacy (including consent and legitimate bases for further processing); etc. Specific challenges faced by forcibly displaced populations: - There has been a proliferation of AI use in border management systems. However, agreed governance standards for AI use in border technologies remain lacking. This has negatively impacted human rights and protection in some contexts, including the ability of people to access territory and seek asylum. - Similarly, the use of AI in (mass) surveillance systems in many contexts contributes to persecution and other human rights violations - There is growing evidence that refugees use AI powered systems to inform movement decisions. In such contexts, there is a proliferation of synthetic content tailored to exploit the vulnerabilities of the populations in question. While governance discussions have rightfully highlighted the need to ensure the safety of children, youth and women, they must also take into account the need for protection of other at-risk populations, such as the forcibly displaced. Opportunities There is a potentially transformative opportunity to ensure that AI governance frameworks take into account the specific realities of forcibly displaced and stateless people. This includes: - Ensuring the inclusion of forcibly displaced and stateless people in targeted capacity-building programmes at the national, regional and local levels (in alignment with paragraph 13(c) of the Global Digital Compact) - The meaningful participation of forcibly displaced and stateless people, and a consideration of their specific needs, circumstances and factors putting them at risk of exclusion or harm, in the design of AI systems, including those utilized in digital public infrastructure.

The Dialogue can play a much-needed role in standards-setting related not as much to the performance of AI systems, as much as to their impact on society. It also has the potential to identify mechanisms to encourage the adoption of interoperable governance frameworks built on certain common, universal foundations. Outputs should include: - A summary of the existing literature in a compact 5-pager that clearly states the expected gains and the known risks, and uses an evidence base to articulate the most pressing need/gap that the Dialogue can address in order to foster a humane AI future. (We need insight, not more long reports.) - A checklist recommending key elements States should include in their governance frameworks. From the perspective of UNHCR, these should include specific measures to ensure inclusion of people with diverse legal status (including refugees, asylum seekers and stateless people within the State's territories) in planning AI infrastructure and to address the specific risks they might face. - Recommendations to AI companies on the governance of their systems (including strengthening of guardrails) to ensure inclusivity, safety and the protection of human rights, including for populations at heightened risk of exclusion or harm. The Dialogue should promote the need for meaningful, long-term engagement with the tech sector. It should also reference the unique role of field-based humanitarian organisations and digital rights organisations to ensure discussions are grounded in realities on the ground, especially in complex emergency settings. Case studies from the field are an important tool for the Dialogue process, bringing abstract technologies and risks to life for policymakers who have limited knowledge of AI technologies and associated risks. Humanitarians can help by contributing more human-centred examples, both positive and negative, for learning and inspiration.

- HLCP and HLCM - Global Refugee Forum Multistakeholder Pledge on Digital Protection: Information Integrity (over 30 commitments including from Member States and leading Digital and AI companies) The Global Dialogue could bring added value by providing a global, multistakeholder forum to establish the normative direction for AI governance both within national frameworks and within tech companies.

We note that the current proposal for the structure of the Dialogue appears to be centred on high-level engagements. We would recommend that provisions be made for the meaningful inclusion of community-based organizations, including refugee-led organizations (RLOs). Measures towards this end might include the following: - We understand that the Co-chairs aimed to identify sources of funding to facilitate the attendance of stakeholders at risk of exclusion. We would advocate for the inclusion of refugee-led and other community-based organizations amongst those benefiting from this support. - With many organizations facing financial constraints, remote participation would also be valuable to ensure a diversity of stakeholder participation/views.

Forcibly displaced and stateless communities represent groups that are consistently at higher risk of exclusion from the benefits of AI, as well as of exposure to AI-driven harms. This includes: - Exclusion from national programming on AI capacity-building and other opportunities - Exclusion from meaningful participation in processes shaping the life cycle of AI technologies - Exposure to harms stemming from targeted disinformation campaigns, in which AI is used to produce, amplify and persuade people with harmful content - Lack of access to protection or rights as a result of the implementation of exclusionary border technologies - Exposure to AI used in conflict or as a modality of persecution (e.g. mass surveillance) However, these groups are currently severely underrepresented in discussions on AI governance. As such, we believe that it is important that special measures be taken to ensure they have clear representation in the Global Dialogue as well as in its outcomes. UNHCR has shared the online submission form with Refugee-led Organizations (RLOs) working in areas related to digital/AI so they can access it if they wish to contribute. However, we would strongly advocate for measures to ensure their representation at the Global Dialogue itself, and would be happy to engage with the Co-chairs of the Dialogue, as well as the Joint Secretariat, on the best way to achieve this.