- Demis Hassabis, Alphabet’s top AI scientist and co-founder of Google DeepMind, held discussions with heads of other major AI labs and Trump administration officials — including Treasury Secretary Scott Bessent — about forming a new independent AI safety entity in the weeks before he stepped down as CEO of Google DeepMind.
- Hassabis has been advancing the idea of an organization that would codify safety guardrails and best practices for developing artificial general intelligence, likening the proposed entity to the International Atomic Energy Agency — a nongovernmental watchdog that promotes cooperation and sets standards on nuclear energy without having regulatory enforcement power over sovereign nations.
- The timing is significant: Hassabis raised the proposal during a period of accelerating AGI capability claims across the industry, with OpenAI, Google, and Anthropic all competing to be the first to achieve human-level or superhuman AI — a race that has raised alarm among researchers about whether safety frameworks can keep pace with capability gains.
- That Hassabis engaged both rival AI lab leaders and the Trump administration signals an attempt to build an unprecedented cross-industry and cross-government coalition around AGI oversight — though the Trump administration’s generally deregulatory posture toward AI makes the political pathway for any formal body uncertain.
What Happened?
In the weeks before relinquishing his role as CEO of Google DeepMind, Demis Hassabis — widely considered one of the pioneers of modern machine learning — held conversations with peers at other AI labs and with Trump administration officials, including Scott Bessent, about creating a new independent industry safety body for artificial general intelligence. Hassabis has framed the proposed entity along the lines of the IAEA: a nongovernmental organization that would establish shared safety standards, facilitate transparency between labs, and codify best practices for AGI development without serving as a formal regulator. The discussions were held in parallel with his transition away from the CEO role, according to people familiar with the matter.
Why It Matters?
An IAEA-style body for AI would represent a fundamental shift in how the industry governs itself — from self-regulation and national AI safety boards toward a coordinated international framework with buy-in from the companies closest to AGI. Hassabis’s credibility as the architect of AlphaFold and one of the most respected technical minds in the field gives the proposal unusual weight; this is not a think-tank recommendation but a pitch from someone who has spent decades working on the systems in question. The inclusion of Trump administration officials in the discussions is particularly notable: Washington has so far favored a light-touch regulatory approach to AI, and any US participation in an international oversight body would require significant political alignment that does not yet exist.
What’s Next?
Watch for whether the proposed entity moves from informal discussions to institutional form — the key variable is whether other frontier labs (OpenAI, Anthropic, Meta AI) publicly endorse the concept and whether any government signals formal participation. If the IAEA analogy holds, the body’s power would ultimately derive from voluntary commitments and reputational pressure rather than legal enforcement, making adoption momentum the critical variable. Hassabis’s post-CEO role at Google DeepMind may give him more flexibility to pursue this as a primary project — a potential indicator that Alphabet is prepared to back the initiative. The proposal will be closely watched by AI safety researchers, policymakers, and investors as a signal of whether the frontier AI industry is moving toward self-imposed structural governance before external regulators mandate it.
Source: The Wall Street Journal
