Hassabis Pushes for Independent AI Safety Body
The DeepMind chief proposes an industry-wide body modeled on the IAEA — and has already pitched it to US officials and rival AI labs.
Illustration · AI-generated (AI IN LIFE)
At a glance
- Proposal: an independent industry body for common AI safety practices
- Conceptual model: the International Atomic Energy Agency (IAEA)
- Talks included Treasury Secretary Scott Bessent and adviser Michael Kratsios
- Builds on Hassabis's July call for a US-led AI standards body
- The US Treasury is reportedly building its own AI oversight in parallel
Demis Hassabis, head of Google DeepMind, is advancing the idea of an independent body for AI safety. The organization would establish common safety practices across the industry as development approaches ever more capable systems. Conceptually, the proposal is modeled on the International Atomic Energy Agency (IAEA).
Hassabis has reportedly discussed the idea well beyond his own company: he spoke with executives at other major AI labs — and with US government officials, including Treasury Secretary Scott Bessent and technology adviser Michael Kratsios. Those conversations suggest the push is more than a thought experiment.
The timing is no accident. Back in July, Hassabis publicly called for a US-led standards body for AI, ideally before year's end. The current push makes that line concrete — while leaning more heavily on an independent, industry-backed model.
The twist is what is happening in parallel: the US Treasury is reportedly building its own oversight approach for AI at the same time. Whether the outcome is a governmental, industry-led, or hybrid model is open — and is likely to become a contest over influence on coming AI regulation.
For companies in Europe, the debate matters because it could create a second, US-shaped governance layer alongside the EU AI Act. Common safety standards among the major labs would effectively apply globally — even without the force of law.
FAQ
What exactly is Hassabis proposing?
An independent, industry-backed body that defines common safety practices for advanced AI systems — conceptually modeled on the IAEA.
How far along is the initiative?
It is a push, not a finished body. Hassabis has talked to other AI labs and US officials, while the US Treasury develops its own oversight plans in parallel.
What does this mean for Europe?
A second, industry-driven governance layer could emerge alongside the EU AI Act. Shared standards among the big labs would effectively reach European users too.
