International Panel on the Information Environment (IPIE)

A successful first Global Dialogue on AI Governance would move beyond general principles toward practical, evidence-based pathways for implementation. In particular, it should: -Establish a shared understanding of how AI systems are already shaping information environments, democratic processes, and societal resilience, grounded in empirical research; -Advance operational approaches to transparency and accountability, including mechanisms for independent evaluation and auditing of AI systems; -Identify concrete steps to improve access to platform-held data for researchers and public-interest oversight, while safeguarding privacy and security; -Foster alignment between governments, industry, and civil society on standards for assessing real-world impacts of AI, especially in high-risk domains such as elections, public health, and climate information. Recent research by the International Panel on the Information Environment (IPIE), including systematic assessments and auditing-focused analyses of generative AI in elections and information ecosystems, demonstrates there is evidence that governance gaps are already materially affecting public discourse and institutional trust. The Dialogue would be most effective if it helps translate such evidence into coordinated international action, including shared benchmarks, policy tools, and ongoing mechanisms for scientific input into governance processes. It must preserve a role for the independent, scientific assessment of AI applications. Significant groundwork already exists to build on. UNESCO's 2021 Recommendation on the Ethics of AI established a global normative framework grounded in human rights and human dignity, and adopted by 193 member states. The OECD AI Principles, first adopted in 2019 and updated in 2024, offer an intergovernmental basis for trustworthy AI, including provisions on transparency, accountability, and robustness. The EU's AI Act demonstrates that risk-based, accountability-sensitive governance frameworks are achievable at scale. The Dialogue should draw on and connect with existing efforts like these, to avoid duplication, but also to address the remaining gaps, particularly around independent scientific assessment, cross-jurisdictional enforcement, and the governance of rapidly evolving frontier systems.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

These priorities reflect the urgent need to align AI governance with observed real-world impacts and the practical conditions required for independent evaluation. Evidence from the IPIE's recent scientific work demonstrates that: -AI systems are already influencing information exposure, amplification dynamics, and public and political discourse, with downstream societal effects; -Existing governance frameworks often lack effective mechanisms for transparency and independent evaluation, limiting accountability; -Access to data, models, and system-level information remains a critical constraint for researchers and public-interest oversight, which needs to be addressed by specific obligations in governance frameworks. Advancing safe and trustworthy AI therefore requires: -Embedding independent auditing and evaluation capacities into governance frameworks; -Ensuring openness that (through open data, open models where appropriate, and interoperable access mechanisms) enables independent scrutiny, and access to data where that openness is not present; -Addressing the systemic incentives and design features that shape AI behavior at scale. These areas are mutually reinforcing. Without sufficient openness and access, transparency and accountability cannot be meaningfully operationalized.

A key cross-cutting issue not fully captured is the need for independent, evidence-based auditing of AI systems and their societal impacts, particularly in the information environment. While transparency and accountability are recognized themes, current approaches often rely on: -self-reporting by developers; -limited and inconsistent access to data and systems for independent researchers. Emerging evidence suggests that this is insufficient to assess how AI systems shape information flows, interact with platform incentives, and influence public understanding and behavior at scale. A critical enabling condition is therefore the development of open and secure access frameworks, including data access regimes and, where appropriate, open-source or inspectable models, which allow meaningful oversight while safeguarding privacy and security. Finally, there is a need to better integrate information integrity as a core dimension of AI governance, recognizing that AI systems increasingly mediate the production, distribution, and visibility of information globally. Addressing these gaps will be critical to ensuring that governance frameworks are grounded in independent, verifiable evidence and capable of adapting to rapidly evolving technologies.

Governance gaps are already producing measurable challenges across regions and sectors, particularly in relation to information integrity. Key challenges include: -Limited capacity and governance frameworks to independently evaluate AI systems, especially generative models deployed at scale; -Insufficient data access and limited openness of systems, constraining robust auditing and independent verification; -Fragmentation of regulatory approaches, leading to inconsistent standards and enforcement; -Rapid deployment of AI tools in sensitive contexts (e.g. elections) without corresponding oversight mechanisms. At the same time, important opportunities are presented: -Advances in AI auditing methodologies and interdisciplinary research provide a foundation for more systematic evaluation frameworks; -Growing recognition among policymakers and International Organisations creates momentum for coordinated governance approaches; -Emerging efforts around open data and access frameworks can enable more effective oversight and collaboration.

International cooperation on AI governance suffers from a common structural problem: the difficulty in translating principles into coordinated, evidence-informed action. Existing frameworks such as the OECD AI Principles, UNESCO's Recommendation on the Ethics of AI, and the EU AI Act provide normative anchors, but no standing body currently bridges independent scientific assessment and intergovernmental policy processes in a systematic way. The AI Dialogue can fill that gap by more clearly articulating a connection and mechanisms of engagement with the Independent International Scientific Panel on AI, and connecting with other independent scientific bodies with demonstrated expertise in assessing AI systems and their societal impacts. Having a structure that brings science to the fore, analogous to the role scientific panels play in climate or biodiversity governance, would give the Dialogue a credible and institutionalized mechanism for grounding international cooperation in evidence rather than aspiration. The IPIE should have a formally recognized role in such structure, to contribute on efforts and avoid duplication. As an independent scientific body already running globally-diverse panels on AI auditing standards, AI and economic inclusion, AI and peacebuilding, AI and elections, and the integrity of AI-shaped information environments, the IPIE is positioned to provide the kind of cross-domain, policy-relevant scientific input that the Dialogue will need on an ongoing basis. This could ensure that independent scientific assessments are structurally embedded in the Dialogue's work, and that scientific work remains responsive to global policymaking needs.

As stated, several existing governance frameworks provide a foundation the AI Dialogue should connect with rather than duplicate. A natural partnership opportunity would lie with the IPIE, as an independent scientific body that produces evidence-based assessments of how information technologies affect public discourse, democratic processes, and epistemic environments through a process of global consensus-building on research outputs. Several of our current scientific panels generate findings directly relevant to AI governance. The Panel on Global Standards for AI Audits is developing criteria and methodologies for evaluating AI systems' public impact, speaking directly to the oversight gap the AI Dialogue seeks to address. The Panel on AI and Economic Inclusion is examining how AI systems affect access to financial services, labor markets, and public resources across different populations. The Panel on AI and Peacebuilding assesses how AI technologies can be deployed responsibly in conflict-affected settings. The Panel on Child Protection and Social Media, and the Panel on Information Integrity about Climate Science, address specific high-stakes domains where AI-driven systems already shape outcomes at scale. A structured partnership between the AI Dialogue, the Independent International Scientific Panel on AI, and the IPIE would allow each initiative to reinforce each other: the IPIE and Independent International Scientific Panel on AI generating an evidence base with the capacity for comprehensive research in connection with scientists who connect it to an even broader scientific network, and the Dialogue translating it into governance-relevant recommendations, and forward-looking exchange informing which questions future panels should prioritize. The AI Dialogue's added value lies in connecting these efforts to reduce fragmentation, elevating evidence-based approaches including AI auditing methodologies, promoting alignment on transparency and access mechanisms that enable meaningful oversight, and providing a neutral space where technical insights meet policy needs.

Different stakeholders bring complementary strengths: -Governments: regulatory authority and implementation capacity; -Industry: technical expertise and system-level insights; -Scientific community and other independent stakeholders: independent research, auditing, and accountability perspectives. However, it is important that the Dialogue goes beyond a space of discussion to be a space for effective policy formulation, recommending and facilitating international agreements that consider the vast evidence that already exists in this space. To leverage these effectively, the Dialogue should combine: -High-level plenaries to align on priorities; -Technical working sessions focused on specific challenges (e.g. auditing, data access, evaluation standards); -Evidence briefings where researchers present findings directly to policymakers. Structured interaction between these groups is essential. In particular, formats that allow direct engagement with empirical research can help ground discussions in real-world evidence. An effective Dialogue would also include mechanisms for continuity between sessions, such as working groups or follow-up processes, to ensure that discussions translate into concrete outcomes. It is also important to strengthen the International Independent Scientific Panel on AI, with a robust and transparent structure and meaningful engagement and partnership with scientific bodies such as the IPIE, to ensure a two-way relationship between policymaking and evidence building

Several perspectives remain underrepresented in global AI governance discussions: -Researchers from the Global South, particularly those studying local impacts of AI on information ecosystems; -Experts working at the intersection of AI, information integrity, and societal resilience; -Civil society actors with direct experience of AI-related harms in elections, conflict, and public discourse. These voices are critical because AI impacts are context-dependent, and governance frameworks must reflect diverse realities. Inclusion can be strengthened through: -Dedicated support for participation and capacity-building; -Mechanisms to integrate regional evidence and case studies into global discussions; -Ensuring that research from diverse contexts informs agenda-setting and decision-making. Broadening participation will enhance both the legitimacy and effectiveness of the Dialogue's outcomes.

To foster meaningful engagement, the Dialogue could adopt formats that prioritize interaction, evidence, and problem-solving: -Policy labs or simulation exercises focused on real-world scenarios (e.g. AI in elections); -Structured evidence sessions, where researchers present auditing findings followed by moderated discussion; -Cross-sector working groups tasked with developing specific governance outputs (e.g. auditing standards, data access principles); -Policy sessions, where experts in different policies are asked to come up with proposals for aligning existing initiatives and identifying the overlaps and gaps; -Iterative consultations, allowing stakeholders to refine proposals over time. Blending technical and policy perspectives in these formats can help bridge gaps between research insights and governance needs.

Recent research on AI and information environments, including by the IPIE, points to a set of concrete governance practices that are already proving effective across contexts. Independent AI auditing with structured data access is increasingly recognized as essential. Research, including work by the IPIE, shows that meaningful evaluation is constrained without access to platform data. Emerging approaches, such as vetted researcher access and secure data environments, are reflected in regulatory frameworks like the EU Digital Services Act and ongoing OECD and UN discussions. Risk-based governance tied to real-world impact domains is also gaining traction. Evidence from AI use in elections, public discourse, and conflict settings demonstrates that risks are unevenly distributed. The EU AI Act provides a leading example of applying tiered obligations based on impact, reinforced by OECD principles. System-level evaluation, rather than a sole focus on model outputs, is increasingly necessary. Research on information integrity highlights how harms emerge from system interactions (e.g. amplification and recommender dynamics). This is reflected in evolving auditing frameworks and discussions within UNESCO and other multilateral fora. Integration of information integrity into governance frameworks is becoming more prominent. IPIE assessments across elections, climate, and peacebuilding show that AI systems shape what information is seen and trusted, an issue increasingly recognized in UN processes. Open and interoperable access frameworks, including standardized reporting and secure research access, are also emerging as good practice, reflected in EU and broader open data initiatives. Finally, structured pathways from scientific evidence to policy, such as scientific panels and advisory mechanisms, are critical. Models like the IPIE and its approach to synthesis research demonstrate how consensus-based research can inform international governance processes. Scaling and connecting these approaches will be essential to ensuring AI governance is operational, evidence-based, and effective in practice.