Global Partners Digital. Also on behalf of: Fundacion Karisma (Colombia), IntenetLab (Brazil), R3D (Mexico), Paradigm Initiative (Nigeria), and Media Foundation for West Africa (Ghana)

What makes the Dialogue unique is its potential to foster multistakeholder dialogue and command high-level political action. To succeed in this goal, the Dialogue must: (1) embrace a truly multistakeholder approach, and (2) adopt a narrow agenda, prioritising a small number of topics that are the most urgent and where there is an existing baseline of agreement to build on. If the Dialogue pursues an overly broad mandate or if it fails to foster genuine multistakeholder dialogue, its impact will be diluted amongst a myriad of other AI governance discussions. In setting this targeted agenda,we strongly advise the Dialogue devote space for: (1) the discussion of recommendations on prohibited AI practices and use cases ('red lines'), building on the existing work of the General Assembly (A/RES/78/265), the OHCHR (A/HRC/60/45), the UN human rights mechanisms (Special Procedures and Treaty Bodies), and the national AI safety institutes; (2) The production of holistic guidelines on human oversight and human-in-the-loop in the context of regulatory oversight; and (3) The development of best practices guidelines across different human rights due diligence and risk mitigation methodologies. While these topics are most closely aligned with thematic clusters 4(e) (respect for and protection and promotion of human rights in the field of AI) and 4(f) (transparency, accountability and robust human oversight of AI systems in a manner that complies with international law), they cut across all thematic clusters and are closely tied to the anticipated work of the AI Independent Scientific Panel. Prioritising the production of practical recommendations and guidelines enhances the role defined for the Dialogue in its mandate and its institutional character inside the UN. These recommendations should be endorsed at the highest political level, guided by research on present-day impacts as well as well-evidenced future risks.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

The three outcomes suggested above for the Dialogue cut across all thematic clusters, including the four clusters highlighted above. In a fragmented AI governance landscape, the Dialogue has a unique opportunity to foster regulatory coherence by advancing a human-rights oriented framework for AI governance, consistent with thematic cluster 4(d). The Dialogue should focus its efforts on topics where there is an existing body of work and agreement which it can build on to incentivise political action. This is true for each of the topics addressed above. The identification of AI 'red lines' and the application of human rights due diligence throughout the AI lifecycle is a fundamental aspect of achieving a rights-based regulatory framework, consistent with thematic cluster 4(e). Advancing AI human rights due diligence will guide the international community by providing for ongoing evaluation, access to remedy and redress squaring the accountability and transparency objectives of cluster 4(e). Extensive work has been conducted by the OHCHR and the UN human rights mechanisms to provide assistance to apply and enforce international human rights law to the internet and digital technologies, including AI. The Dialogue should draw renewed political attention to this work and build on it. The consistent application of AI 'red lines' and human oversight is also an essential component of thematic cluster 4(a). Amid increased reports of agentic AI tools behaving erratically, there is a clear need for guidance on what constitute essential security features to prevent unacceptable model behaviours and outcomes. Additional guidance outlining the necessary elements of human oversight and recommending the use of tools such as model cards to report on AI system capacities is also essential. Our hope is that these elements would support a solid baseline for a cross-border oversight that advances AI transparency and accountability.

Two topics that are not explicitly addressed in the list above but that cut across many themes are the environmental impact of AI-related infrastructure such as hyperscaler infrastructure and labour protections for the workforces engaged in AI development. As mentioned above, we advise that the Dialogue adopts a narrow agenda, prioritising for thematic discussion a small number of topics that are the most urgent and where there is an existing baseline of agreement to build on. However, the topics highlighted here are of great importance and should be integrated into the broader discussions during the high-level plenary segments. Discussion of AI's physical infrastructure and environmental impacts of it is vital to understand the AI's social, economic and ethical implications, in particular in terms of its negative impacts on human rights, such as the right to an adequate standard of living for communities impacted by AI hyperscaler development and the impacts on labour and production systems that have the potential to exacerbate poverty and inequality in systemic manner. The impacts of AI in labourforce have already been evidenced by numerous, extensively reported cases of large AI companies outsourcing tasks such as AI data labelling and content moderation to external companies which fail to provide adequate pay or working conditions. These topics of discussion are critical to identifying and remedying the human and societal costs of the increasing adoption of AI systems. They will require dedicated space in the Dialogue in the future as the evidence of impacts is collected and offered for discussion.

Across the four themes we identified, an overarching gap is the weak or absent adoption of human rights-oriented regulatory frameworks related to AI. The result of this is that AI has been designed, developed and deployed across jurisdictions in a manner that does not comply with existing international human rights law obligations. At a minimum, the adoption of a human rights-compliant regulatory framework for AI requires the integration of human rights due diligence, including ex ante and post hoc human rights risk assessment. Human rights risk assessment and risk mitigation should apply throughout the AI lifecycle, encompassing both AI-related technologies (such as foundation models) and specific applications. There should be ongoing evaluation to identify and prohibit unacceptable technologies or applications. In the absence of these frameworks, we have observed the ongoing development and deployment of AI systems that pose unacceptable risk to human rights, including present-day risks, such as mass surveillance (including biometric surveillance), crime prediction systems, social scoring, as well as well-evidenced future risks, such as the development of AI systems that autonomously commission cyber attacks or assist weapons development. The enforcement of a human rights-respecting regulatory framework also requires guarantees of transparency, ongoing evaluation, and accessible and effective remedy and redress mechanisms. In the absence of this, we have observed that whole populations have not been notified where they have been subject to AI-driven decision-making and have been unable to access remedy or redress for AI-related harms. This includes real-world harms such as: the spread of non-consensual intimate images (NCII) being used also to harm democracy by silencing women activists, human rights defenders, journalists; deepfakes; discriminatory decision-making following the adoption of AI systems in social security, healthcare; policing and border security; mass copyright and privacy violations caused by data scraping; and, risks related to information integrity. While acknowledging this significant challenge, we have observed an important opportunity in the growing development of human rights-oriented regulation and guidance. This provides a significant baseline of work for the Dialogue to refer to and build on. This is discussed further under question 18.

As earlier mentioned, the central function of the Global Dialogue on AI Governance should be to foster multistakeholder dialogue and catalyse foreign policy cooperation on AI governance. We strongly advise that the Dialogue prioritises for thematic discussion a small number of the most urgent topics where there is an existing baseline of agreement to build on. This approach is most effective in allowing the Dialogue to build on the work of existing AI governance initiatives, while complementing this with the higher level status and significant political attention which a Dialogue format can offer. It is important that the Dialogue also prioritises those topics which connect closely to the work of the AI Independent Scientific Panel, providing a multistakeholder format for the further discussion and debate of the results of the Panel's annual reports. It is with these considerations in mind that we propose three outcomes for the first phase of the dialogue's work. These are: (1) the production of further recommendations on prohibited AI practices and use cases ('red lines'), (2) the production of holistic guidelines on human oversight and human-in-the-loop in the context of regulatory oversight, and (3) the production of guidelines and best practices on human rights due diligence and human rights risk assessment. The Dialogue should leverage in this work multistakeholder and multidisciplinary approaches built into its mandate and nurtured by the close collaboration with the Panel, allowing its recommendations to be comprehensive of both present-day harms and well-evidenced future risks. The Dialogue should aim to produce outcomes on these topics to be offered for endorsement to the UN General Assembly. This would support international consensus building at the highest political level.

To ensure coordination and reduce duplication across the UN system which undermines the capacity of diverse actors to engage, it is essential that the Dialogue builds on and connects with other mechanisms and processes established under the auspices of the World Summit of the Information Society (WSIS) and the GDC. Chief amongst these is the Internet Governance Forum (IGF) and its wider ecosystem of National and Regional Initiatives (NRIs) as well as the ongoing work of the CSTD Working Group on Data Governance. In 2026, the global IGF is ideally timed at the mid-point of the first and second AI Dialogues. We strongly advise that the co-facilitators of the Dialogue take advantage of this opportunity, making use of the IGF as an opportunity for mid-point evaluation of the outcomes of the first Dialogue, and to identify priorities for discussion in the second Dialogue. We also advise the co-facilitators to work with the IGF Secretariat and the IGF MAG to explore how the Dialogue can make use of the IGF's intersessional workstreams as a way to progress the high-level outcomes agreed at the Dialogue through a multistakeholder approach. Otherwise, the Dialogue risks being siloed, time-limited discussions, with little capacity to act on agreed discussions. As discussed above, the Dialogue should also establish clear connections with the ongoing work of UN agencies and entities. Specifically, we highlight the extensive work of the OHCHR and the UN human rights mechanisms to apply and support the effective enforcement of international human rights law in the context of AI governance. Close connection with the OHCHR and mechanisms will be essential to achieving progress in thematic cluster 4(e).

There are a number of valuable resources which can be referred to. The NETMundial+10 outcome document proposes guidelines to mitigate power asymmetries in the context of multistakeholder and multilateral decision-making processes. The 'Five-Point Plan for an Inclusive WSIS+20 review', developed by a cross-stakeholder community of over 100 groups and experts, offers practical guidance in the context of a multilateral process. In addition, we make the following recommendations: (1) Publishing clear timelines for input and milestones and providing sufficient time for stakeholders to submit input in different formats. (2) Maintain a centralised repository of all input received and draft documents, their status and stakeholders consulted. Publish substantive summaries and recordings of the Dialogue and all preparatory meetings. (3) Ensure diverse participation by applying inclusive criteria to any accreditation processes. Including facilitating remote participation for stakeholders that cannot participate in person. (4) Facilitate meaningful stakeholder consultations, leveraging existing community convenings to host 'satellite' consultations, building on positive practice during the WSIS+20 review process. Provide opportunities for regional and cross regional consultations. (5) Structure sessions to allow non-governmental and governmental stakeholders to respond to each other in real time, showcasing diversity of approaches towards one selected topic rather than isolated interventions. (6) Prioritise formats that facilitate consensus-building and agreement of clear outcomes. This includes incorporating a dedicated segment focused on summarising outcomes and follow-up with other AI and human rights related forums. (7) Appoint a non-governmental stakeholder liaison to support the Co-Chairs to facilitate meaningful multistakeholder engagement ahead of, during, and following the AI Dialogue. Consider leveraging the support of this group when drafting the Co-Chair summary outcome of the AI Dialogue. (8) Make use of the IGF's existing intersessional workstreams to bring in a wide range of expertise to inform the Dialogue's outcomes. This should include those with lived experience of AI-related harms.

Global discussions on AI governance must ensure meaningful participation of civil society, women human rights defenders, LGBTQIA+ rights advocates, labour organisations, and communities in the Global Majority across the entire AI policy cycle, including from agenda setting to design and deployment, as well as evaluation and assessment frameworks. The communities who are most frequently excluded or underrepresented in global discussions on AI governance are those most negatively affected by AI development, deployment, use and adoption. This includes workers throughout the AI value chain, such as low-paid data annotators and content moderators. These workers face exploitation, poor working conditions, insecure work without adequate labour or social protections, and are principally from countries in the global Majority. It includes individuals and groups negatively impacted by the use of AI systems in welfare, policing, border control and other sectors where these systems are trained on limited or biased data or lack adequate safeguards. It also includes communities impacted by AI hyperscaler development. Concerted effort should be made to engage affected communities and provide necessary accommodation to enable them to participate fully in the Dialogue. Their participation is necessary to understand the ethical, social, economic, ecological and political impacts, harms, inequalities and risks raised by AI, and to ensure that discussion of policy solutions are practically implementable, grounded in present-day use cases, and have the legitimacy to be implementable.

While it is important to adopt innovative engagement formats, it is equally important to use tried and tested methods to effectively engage with stakeholders and facilitate the right to participation. We recommend the following measures: (1) Publishing clear timelines for input and milestones and providing sufficient time for stakeholders to submit input in different formats. (2) Maintain a centralised repository of all input received and draft documents, their status and stakeholders consulted. Publish substantive summaries and recordings of the Dialogue and all preparatory meetings. (3) Ensure diverse participation by applying inclusive criteria to any accreditation processes. Including facilitating remote participation for stakeholders that cannot participate in person. (4) Facilitate meaningful stakeholder consultations, leveraging existing community convenings to host 'satellite' consultations, building on positive practice during the WSIS+20 review process. Provide opportunities for regional and cross regional consultations. (5) Structure sessions to allow non-governmental and governmental stakeholders to respond to each other in real time, showcasing diversity of approaches towards one selected topic rather than isolated interventions. (6) Prioritise formats that facilitate consensus-building and agreement of clear outcomes. This includes incorporating a dedicated segment focused on summarising outcomes and follow-up with other AI and human rights related forums. (7) Appoint a non-governmental stakeholder liaison to support the Co-Chairs to facilitate meaningful multistakeholder engagement ahead of, during, and following the AI Dialogue. Consider leveraging the support of this group when drafting the Co-Chair summary outcome of the AI Dialogue. (8) Make use of the IGF's existing intersessional workstreams to bring in a wide range of expertise to inform the Dialogue's outcomes. This should include those with lived experience of AI-related harms.

As earlier mentioned, extensive work has been conducted by the OHCHR and the UN human rights mechanisms (UN Special Procedures and the treaty body system) to provide assistance to apply and enforce international human rights law to the internet and digital technologies, including AI. The General Assembly and the UN human rights system have produced specific guidance related to the prohibition of unacceptable AI applications. The Dialogue has the opportunity to draw renewed political attention to this work and build on it. The Dialogue should also benefit from the considerable, previous work undertaken by other multilateral bodies. Particular attention should be paid to the efforts of the Council of Europe to develop a human rights, democracy and rule of law-oriented regulatory framework for AI, one of the few explicitly rights-based regulations in this field. Particularly its methodology for the Risk and Impact Assessment of Artificial Intelligence Systems from the Point of View of Human Rights, Democracy and Rule of Law (HUDERIA). The risk-based, market-regulation approach to AI regulation of the European Union (EU) is also notable. The work of the OECD and the Global Partnership on AI (GPAI), in particular through the development of the OECD Recommendation on AI, have produced relevant expertise, including the elaboration of a definition of AI which is widely used, including in the EU's AI Act. The G7 reporting framework - Hiroshima AI Process (HAIP) international code of conduct for organizations developing advanced AI systems. And the just released Due Diligence Guidance for Responsible AI which contains a 6-step framework for identifying and mitigating AI-related adverse impacts, endorsed by all OECD members plus 17 partner governments. The unique value of the Dialogue is the opportunity it provides to build coherence across UN and non UN AI governance processes to prevent normative fragmentation and support consensus building for a positive agenda on AI accountability.