DeepMind CEO Hassabis calls for independent AI standards body modeled after CERN and IPCC

DeepMind CEO Demis Hassabis called for the creation of an independent international standards body to oversee frontier AI development during his keynote at the AI Summit London 2026. Hassabis argued that voluntary industry commitments and fragmented national regulations are no longer adequate to manage the risks of increasingly capable AI systems.

"We need a CERN or IPCC for AI — a neutral, expert-led organization that can set technical standards, conduct independent evaluations, and provide early warnings about emergent capabilities," Hassabis said, drawing parallels to the International Atomic Energy Agency and the Intergovernmental Panel on Climate Change.

The proposed body would focus on "frontier models" — those trained on massive compute with capabilities posing systemic risks — and would serve three main functions: developing standardized tests for dangerous capabilities like autonomous replication and cyberattack assistance, creating a global AI incident database, and requiring developers of the largest models to submit safety cases before deployment, similar to FDA drug review processes.

Hassabis acknowledged that DeepMind, itself a leading frontier model developer under Alphabet, would be subject to the same rules. "This isn't about regulating others — it's about creating a level playing field where safety is a prerequisite, not an afterthought," he said.

The proposal comes amid growing international concern about AI safety. The UK government has announced a second global AI safety summit for later this year, and the EU's AI Act is scheduled to take full effect in 2027. However, Hassabis argued that regulatory fragmentation could allow dangerous models to slip through gaps. "No single country can police AI — the models are digital, they are global, and they are improving faster than any one government can track," he said.

Several other AI leaders have voiced similar calls. Last month, OpenAI CEO Sam Altman proposed an "AI safety inspectorate" in a Washington Post op-ed, and researchers including Yoshua Bengio and Geoffrey Hinton published a paper calling for mandatory licensing of AI developers.

Critics questioned the feasibility of such a body, citing the rapid pace of technological change and the concentration of AI expertise in private companies. "The governments would need to attract top AI talent away from industry, and that's incredibly expensive," said Dr. Emma Richards, a policy fellow at the Ada Lovelace Institute. "There's also the risk that standards become a barrier to entry for smaller players."

Hassabis acknowledged these challenges but argued the alternative — a race to the bottom on safety — is worse. He called on governments including the US, China, and the EU to jointly fund the organization with an estimated $2-5 billion per year. "We have the chance to get this right before the most dangerous capabilities arrive," he said. "Let's not wait for a catastrophe to act."