Google DeepMind CEO Demis Hassabis called on July 14 for the creation of an independent “Standards Body” to test the world’s most advanced AI systems before they reach the public, arguing that artificial general intelligence is now “probably only a few short years away.”
In a long-form essay titled “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis proposed a US-led organization modeled on the Financial Industry Regulatory Authority (FINRA) — a federally overseen but industry-funded self-regulatory body, rather than a direct government regulator.
How the framework would work
Under the plan, frontier labs would initially share unreleased models with the Standards Body voluntarily, up to 30 days before launch. Models crossing certain capability thresholds would be classified “Frontier-class” and undergo evaluations covering cybersecurity, biological-weapons risk, deceptive behavior and autonomous agentic capability, with benchmarks updated quarterly. Hassabis said the body should eventually run independent, held-out tests so labs cannot simply train to pass the public evaluations.
Funding, he wrote, “would need to be substantial and likely mostly come from industry,” with governance handled by a board including independent technical experts and open-source representatives alongside government and industry seats. Voluntary participation could later be formalized into a requirement once the testing regime proves “effective and robust.”
Why now
Hassabis grounded the proposal in escalating risk: as models grow more agentic and capable of recursive self-improvement, he wrote, maintaining human control becomes harder, while dangers tied to cyberattacks and nuclear and biological weapons design grow more acute.
The proposal also responds to criticism of existing US review processes. According to TechCrunch, earlier ad hoc government reviews of Anthropic’s Mythos and OpenAI’s Sol models drew criticism for lacking technical expertise and operating with little transparency — a gap Hassabis’s framework is designed to close without imposing direct government regulation.
The idea faces political headwinds. White House AI advisor Sriram Krishnan has said there “will not be an FDA for AI,” reflecting resistance within the Trump administration to formal oversight structures, even ones that are industry-funded and self-regulatory by design.