Several recently published, open letters have called for stronger oversight of how artificial intelligence (AI) is developed and deployed. Looking at these letters side-by-side reveals a shared set of high-level concerns articulated by a wide range of diverse individuals and institutions engaged in various forms of AI governance. Yet the fact that these efforts were organized more-or-less simultaneously but independently, the relatively small number of overlapping signatories, and their different substantive priorities also reveal a critical lack of stakeholder coordination and coherence across the field.
One of these letters – Centering Human Rights in AI Governance (“Centering Human Rights”) – was issued in October by the Multistakeholder Approaches to Participation in AI Governance (MAP-AI) project and was signed by 84 primarily global majority-based academics, civil society organizations, and technical experts. MAP-AI is co-led by the Centre for Communication Governance at NLU Delhi and the Global Network Initiative, the organization that I direct. In order to understand how the views of MAP-AI’s diverse but primarily Global Majority-based community compare with those of other actors calling for greater scrutiny and oversight of AI, I compared it with the following:
In line with MAP-AI’s focus on building and fostering collaboration across multistakeholder communities engaged in AI governance, we see these letters as an encouraging opportunity to continue building partnerships between key actors who share a commitment to fostering AI governance that is inline with core democratic and human rights principles.
Despite their different origins and audiences, the four statements have a surprising amount in common. All of them are framed at least in part as responding to the escalating pace of AI capabilities and associated risks. Pacing the Frontier focuses most explicitly on the competitive dynamics that it cites as driving companies and countries to pursue new capabilities and hindering related safety and coordination measures. Centering Human Rights also notes how the pace at which AI is developed and deployed is “driven by intense industry and geopolitical competition,” highlighting the resulting exacerbation of existing social and economic inequities and gaps in regulatory capacity. Meanwhile, Minimum Conditions and Trust Needs Evidence are positioned as responses to the AI industry’s own, increasingly urgent calls for external scrutiny and support.
All four also recognize independent oversight as an essential element of governance. While Trust Needs Evidence and Minimum Conditions are focused on the role of third-party evaluation, Pacing the Frontier calls explicitly on the U.S. government to help “buy time to address emerging risks, develop security measures, and strengthen oversight.” Centering Human Rights takes an even broader approach, calling on all stakeholders to engage urgently in “inclusive, coordinated, and rights-centered efforts to build trust, enhance transparency, and ensure oversight and accountability.”
Three of the statements point to the critical need for shared standards. Trust Needs Evidence calls for international comparability through shared standards, and the Minimum Conditions letter says conditions for credible evaluation should be increasingly standardised, codified and enforced. The Center Human Rights letter notes that assessments using frameworks developed together with companies, civil society, academics, and the technical community are more effective than “ad hoc, bespoke efforts.” The same three letters also ask for transparency in how evaluation and/or governance are developed, conducted, and reported.
Significantly, three of the four letters explicitly recognize the need for international coordination. Center Human Rights calls on governments to “pool resources, share lessons, and build common tools;” Trust Needs Evidence wants results comparable across borders; and Pacing the Frontier asks the US government to support an “international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”
Notwithstanding the above-mentioned similarities, the four letters also display some important differences in terms of their framing, focus, audience, and signatories.
Pacing the Frontier is explicitly focused on the specific risk that, as AI research is automated (i.e., through recursive self-improvement), the technologies’ capabilities could accelerate beyond humanity’s ability to understand and control them. While the other three letters implicitly or explicitly recognize this macro risk, they also foreground present-day harms to varying degrees. Centering Human Rights addresses this most explicitly, calling on stakeholders to “build governance focused on today’s lived realities, alongside tomorrow’s technological frontier” and identifying international human rights law explicitly as the “essential floor” upon which to build governance. Trust Needs Evidence calls for measuring what matters to the people affected by AI systems, including real-world effects after deployment, informed by civil society and affected people, and Minimum Conditions includes significant incidents of real-world harm within evaluators’ scope. These last two letters focus exclusively on process principles (i.e., access, independence, transparency), without addressing the substantive foundations upon which evaluation standards should be built.
In terms of audience, Pacing the Frontier is addressed directly to the US Government, while Minimum Conditions appears to be oriented more toward the leading, mostly US-based AI companies that are calling for and would ultimately be the subject of independent evaluations. Trust Needs Evidence calls on “governments, funders, developers, and standards bodies to create the conditions for independent AI evaluation that is rigorous, reliable, and credible to the public.” Centering Human Rights again takes the broadest approach, making recommendations to governments and companies around the world.
Finally, the letters’ respective signatories reveal differences in approach that exist within the broader AI governance ecosystem. Trust Needs Evidence is signed by individuals in a personal capacity and is the most cross-sector, led by academics alongside civil society, former government officials, and some representatives of standards bodies and AI companies. The Minimum Conditions letter is also signed by individuals, though its signatories are drawn more heavily from the community of third-party evaluators, AI safety nonprofits, and academics. Pacing the Frontier has by far the largest and least diverse group of signatories, featuring over a thousand named individuals (with hundreds more “anonymous” signatures) working in or with frontier AI companies. About half of the signatories to Centering Human Rights are individuals, with the other half representing civil society groups and academic institutions from the digital rights community, so each signature often represents a wider constituency.
Geographically, Centering Human Rights is by far the most diverse, featuring signatories from East, West, Southern and North Africa, South Asia, Southeast Asia, the Middle East, Latin America, Taiwan, Australia, Europe and North America. As far as we could tell from published affiliations, the signatories to the other three letters are almost exclusively North American, with a smaller number from Europe and only a handful from outside these “Western” regions. This highlights an important gap in the discourse, which MAP-AI is actively working to address. Across all four letters we found limited, but important overlap in signatories. Not surprisingly, the most overlap exists between the two letters focused on evaluation – Minimum Conditions and Trust Needs Evidence, while there was zero overlap between Pacing the Frontier and Centering Human Rights. This highlights both the siloed nature of stakeholder communities in the AI ecosystem and important convergence on high-level priorities for AI governance across stakeholder groups.
These four statements come from different communities with different roles in the AI ecosystem, each with its own forms of power and leverage: frontier AI company employees, independent evaluators, academics and former officials, and civil society organisations across the Global Majority. They speak to different audiences and propose different mechanisms, but they agree on more than these differences might suggest.
All four letters identify this moment as a critical opportunity for embedding and standardizing oversight, enhancing transparency, and fostering accountability around the development and deployment of AI. And, critically, they all recognize the fact that this cannot be done by any one company, country, or community alone. These points of convergence provide a foundation for shared approaches to AI governance.
The differences are just as useful. They show where the conversation is still unsettled: whether governance should focus on frontier risk, present-day harms, or both; whether coordination should run through a single international effort or a more distributed ecosystem; and how far governments should act, and with what safeguards.
The signatory lists make the same point. Each statement draws on a different community, and few people appear in more than one. The geographical representation of the signatory list also shows who is missing from the conversation, with a lack of representation from stakeholders in the Global Majority standing out as a gap. This points to the need for more inclusive and diverse discourse to continue to build legitimacy around shared priorities and meaningful coordination. Bringing these communities into closer conversation, including people whose perspectives are less visible in current debates, would strengthen all of these efforts.