Google DeepMind Chief Urges Urgent AI Oversight Body Before AGI Arrives

Google DeepMind CEO Demis Hassabis has called on the US government to urgently form a new oversight body for advanced AI models — saying the system should become operational "within months," ideally before the end of 2026, while the industry still hasn't reached AGI (artificial general intelligence).

What Hassabis actually proposed

The proposal appears in Hassabis's personal manifesto, "A Framework for Frontier AI and the Dawning of a New Age," published Tuesday, July 14, 2026. Hassabis is proposing a body modeled on FINRA (the Financial Industry Regulatory Authority), a US nonprofit funded by the industry it regulates but operating under government oversight — in FINRA's case, under the Securities and Exchange Commission.

How the model testing would work

In the initial, voluntary phase, frontier AI labs would submit their models for testing 30 days before public release. The tests are meant to catch dangerous capabilities in cybersecurity and biological research, as well as signs of "deception" — attempts by models to bypass built-in safety guardrails. The framework also covers watermarking of AI-generated content. Evaluation benchmarks would be refreshed quarterly, and faster over time.

Once the testing protocol proves "effective and robust," the voluntary regime could quickly shift to a mandatory one — meaning frontier models would need to pass evaluation just to enter the US market at all. The rule would apply regardless of a model's country of origin or whether it's open- or closed-source, with an exemption only for non-frontier models from startups and academic research.

Who would run the new body

Under Hassabis's plan, the governing board would include independent leading technical experts — including Turing Award recipients — along with representatives from the open-source community, government, and industry. Hassabis said his talks with the Trump administration on the proposal have already produced positive signals.

Why now

Hassabis estimates AGI — a system with all the cognitive capabilities of the human brain — is "probably only a few short years away." His initiative is already the second of its kind in the past month: roughly a month earlier, Anthropic CEO Dario Amodei proposed a similar oversight framework modeled instead on the FAA, the US aviation regulator. Two leaders of major AI labs independently calling for systematic oversight, in such a short window, is itself a notable shift in tone from an industry that until recently was far more guarded about external regulation.

For more on how Google DeepMind itself came together and how Gemini fits in, see our full breakdown: Gemini and Google DeepMind: The Full Story

What this means in practice

For now, Hassabis's proposal is a manifesto, not law: even the voluntary version of the system needs buy-in from major labs and support from the US government to actually launch. But the fact that two of the industry's biggest players independently proposed similar oversight mechanisms within a single month suggests mandatory pre-release model testing is moving from theoretical debate toward a concrete initiative already being discussed with government.