Sources: Demis Hassabis pitched a new independent industry AI safety entity, modeled on the IAEA, to top Trump officials before stepping down as DeepMind CEO (Wall Street Journal)

Demis Hassabis, CEO of Google DeepMind, engaged in discussions with officials from the Trump administration and leaders of other AI companies in the months leading up to DeepMind's management change. According to the Wall Street Journal, Hassabis proposed the formation of an independent industry safety body to establish safety guardrails and best practices for developing artificial general intelligence (AGI).

Comparing his idea to the International Atomic Energy Agency, Hassabis aimed for the organization to function similarly to a watchdog for AI.

During a visit to Washington earlier this summer, Hassabis met with Treasury Secretary Scott Bessent and tech adviser Michael Kratsios to discuss his plan. Subsequently, Hassabis published an X essay advocating for a U.S.-led, industry-funded standards body to evaluate AI models, akin to the Financial Industry Regulatory Authority. Under this proposal, federal agencies and national laboratories would collaborate with AI companies to assess models pertinent to national security and identify "frontier-class" systems.

While the powerful AI labs are racing toward AGI, concerns about potential risks have been voiced by leaders such as Hassabis, Sam Altman of OpenAI, and Dario Amodei of Anthropic. Altman recently proposed a U.S.-led international forum, and Anthropic supported a global agreement to delay development and verify compliance. Hassabis's efforts for safety mechanisms represent a notable step by a leading lab official in establishing such mechanisms.

Recently, Google announced a management overhaul at DeepMind, with Hassabis stepping down as CEO but becoming chairman of DeepMind and chief scientist of Alphabet. In a memo, Hassabis expressed his belief that AGI is near and decided it was time to shift focus. Alphabet CEO Sundar Pichai stated that Hassabis's new role allows him to concentrate on shaping the future of AGI.

Critics argue that voluntary and self-regulatory approaches are not sufficient to tackle AI's security and economic risks, calling for mandatory model reviews and stronger government oversight. The Trump administration has heightened its scrutiny of AI's security risks but maintains voluntary evaluations, leaving the debate over regulation unresolved as AGI development progresses.