Hassabis Pushes for Independent AI Regulatory Body Before Stepping Down as DeepMind CEO
Nashnova编辑部
Google DeepMind co-founder Demis Hassabis spent weeks lobbying U.S. officials and rival lab heads for an industry-funded AI safety standards body modeled on the IAEA — the most concrete regulatory push yet by a leading AI lab executive, coming just before he handed the CEO role to his deputy.
What kind of body is he proposing?
Hassabis compared his proposed agency to the International Atomic Energy Agency (IAEA) — a non-governmental body that promotes nuclear cooperation while enforcing safety oversight. This means → he envisions industry self-funding and self-regulation under external scrutiny, not direct government control of AI.
His more specific template: FINRA — the Financial Industry Regulatory Authority, a broker-dealer self-regulatory organization dating back to the Great Depression. In plain terms = the finance industry already runs a "police yourself, but with teeth" model; Hassabis wants the same logic for AI.
The plan's core: U.S. federal agencies and Department of Energy national labs would partner with AI companies to test models in national-security-relevant domains and establish a "Frontier-class" certification standard.
Who did he brief before going public?
Weeks before posting on X, Hassabis pre-briefed Treasury Secretary Scott Bessent and Trump's chief technology adviser Michael Kratsios. This means → the July post was not spontaneous commentary — it was a choreographed policy push.
Key context: Bessent had previously warned that a recent Anthropic model poses a cybersecurity threat to the financial system. This reflects a shift inside Washington — AI risk anxiety has moved from a long-horizon hypothetical to an immediate policy concern.
The Trump administration has stepped up scrutiny of AI safety risks, yet evaluation of the most powerful models remains voluntary and no regulatory framework has taken shape.
Why did his own role change at this exact moment?
Last week Google announced: Hassabis stepped down as DeepMind CEO, replaced by deputy Koray Kavukcuoglu. Hassabis becomes DeepMind chairman and Alphabet chief scientist.
In an internal memo he wrote: "I have worked my whole life toward AGI, and now, like many of you, I feel it is within reach." In plain terms = he believes general AI is no longer a distant goal — and precisely because of that, safety rules must be set now.
This reflects a subtle timing window: the moment of stepping up from CEO to a broader role is the ideal launchpad for an industry-wide safety agenda — no longer representing one company's daily operations, but speaking as a scientist about the entire field's governance.
What are the other AI giants saying?
OpenAI CEO Sam Altman last month proposed a "U.S.-led international forum" to set standards and govern labs.
Anthropic backs a global agreement to slow development and verify compliance.
Hassabis's proposal is considered the most specific and actionable of the three: not just calling for forums or treaties, but sketching out institutional architecture — who funds it, who runs the tests, what the certification standard looks like. This means → the AI governance race has moved past "whether to regulate" into "how to regulate, and who gets to set the rules."
Content is for reference only, not financial advice.