Former Leaders Urge International AI Agency to Mitigate Existential Risks

A paralysis in governing advanced artificial intelligence is causing present harm and increasing the likelihood of future catastrophe, according to an opinion piece co-authored by former UN High Commissioner for Human Rights Mary Robinson and former Director-General of the Organisation for the Prohibition of Chemical Weapons Ahmet Üzümcü. They argue that narratives excusing inaction—that the technology moves too fast for oversight, that companies can self-regulate, or that geopolitical rivalry should take precedence—are dangerous.

While acknowledging the unknowns surrounding the most powerful AI systems, the authors emphasize that AI's risks stem from human intelligence and necessitate stringent guardrails. These risks transcend national borders, rendering unilateral action insufficient. The authors assert that waiting for complete information or a softening of great power competition is not an option, and international cooperation on AI safety is urgently required.

Robinson and Üzümcü draw upon decades of experience in global risk management, citing precedents like the Montreal Protocol for ozone layer protection, the smallpox eradication campaign, the Nuclear Non-Proliferation Treaty (NPT), and the Chemical Weapons Convention. These efforts, they note, demonstrated that international coordination on existential threats is possible even in the absence of trust, proving that rivals can recognize dangers greater than their rivalries.

Lessons from Past Treaties

The NPT's framework, which involved states forgoing dangerous capabilities in exchange for access to technology and a voice in governance, offers a potential template for AI. Access to computing power and models could be traded for compliance, mirroring the exchange of enriched uranium and reactor technology in the past. The Chemical Weapons Convention, with its near-universal membership and a neutral verification body like the OPCW, provides another instructive model for building confidence through independent inspection.

For AI companies, a binding and verifiable safety floor is presented not as a threat, but as a crucial assurance. It offers credible proof to the public and competitors that the development of increasingly powerful AI systems does not come at the cost of cutting corners on safety.

Proposing an International AI Agency

While no single historical model is a perfect fit for AI, the authors propose the establishment of an international agency for frontier AI. This agency would be tasked with verifying that the world's most powerful AI systems are safe, remain under human control, and help prevent international crises. It would serve as a hub for reporting and monitoring serious AI incidents and near misses across borders, allowing failures in one entity to serve as warnings for the global community. The agency would also facilitate the sharing of knowledge about AI capabilities and risks between governments and companies. Crucially, it would possess the authority to independently verify safety and, where safety cannot be demonstrated against agreed-upon "red lines," to mandate a pause in development.

The authors frame this proposal not as a call for panic, but as an appeal for the political courage and leadership exemplified by past international agreements. They warn against waiting for an "AI Chernobyl moment" and urge world leaders to commit to negotiating such an international agency to address the existential risks posed by advanced AI.