All three major AI lab CEOs have now published near-identical calls for oversight in five weeks. Hassabis wants it running before the government's own August 1 framework has even had time to prove itself, worth understanding as two parallel oversight tracks rather than one.
- Demis Hassabis, Google DeepMind’s CEO, published a manifesto calling for a US-led AI standards body modeled on Wall Street regulator FINRA.
- The proposed body would run 30-day pre-release reviews of frontier models, checking for dangerous capabilities, safety bypasses, and deceptive behavior.
- Hassabis wants it operational before the end of 2026, starting voluntary before potentially becoming mandatory for any frontier model entering the US market.
- OpenAI’s Sam Altman and Anthropic’s Dario Amodei have published similar proposals in the past five weeks, though Amodei wants a federal agency with power to halt launches outright.
- This follows the US government’s restriction of Anthropic’s Mythos models over safety bypass concerns earlier this year.
What Happened
Demis Hassabis, Google DeepMind’s CEO and a Nobel laureate, published a manifesto this week calling for a US-led AI standards body modeled on FINRA, the private, industry-funded regulator that oversees Wall Street under SEC supervision. He argued frontier AI’s cybersecurity risks are “warning shots” for more serious threats ahead.
The proposed body would feature a majority-independent board combining technical experts, government officials, and open-source representatives. Frontier labs would initially submit models voluntarily for a 30-day pre-release review, checking for dangerous autonomous capabilities, safety bypass attempts, and deceptive behavior, before potentially becoming mandatory for any frontier model entering the US market.
Google senior director Owen Larter said AI companies should contribute technical expertise, but their board influence should stay “proportionate and balanced” against other members. Hassabis wants the body operational before the end of 2026 and says the rules should apply equally regardless of a model’s country of origin or open versus closed license.
This follows the US government’s restriction of Anthropic’s Mythos models earlier this year over concerns its safety controls could be bypassed. Sam Altman and Anthropic’s Dario Amodei have each published similar oversight proposals in the past five weeks, though Amodei’s version calls for a federal agency with power to halt model launches outright, a more aggressive stance than Hassabis’s industry-led model.
Why It Matters
Three competing AI lab CEOs converging on nearly identical oversight proposals within five weeks is a stronger signal than any single company’s position. It suggests the industry sees a real choice ahead: shape the rules themselves now, or have stricter rules imposed on them later.
This proposal doesn’t replace the government’s own August 1 access framework, it runs alongside it. Hassabis is pitching an industry-funded, FINRA-style body as a complement to, not a substitute for, the direct government review process already underway. Whether these two oversight tracks end up aligned or in tension is exactly the kind of detail that will determine how predictable AI model access becomes for anyone building on it.
“Other threats including nuclear and bio risks may soon emerge as capabilities continue to advance.” Demis Hassabis, CEO, Google DeepMind
Bottom Line
Watch whether the Trump administration formally engages with Hassabis’s proposal, and whether Amodei’s more aggressive halt-authority version gains any traction against it. The gap between these two positions is where the real policy fight will play out.
For SaaS founders, this is another data point confirming AI governance is now a genuine vendor risk category, not a settled question. Keep tracking how the August 1 framework and any industry-led standards body interact, since together they’ll shape how predictable your AI vendor’s model releases are going forward.
