DeepMind CEO Proposes AI Guardrails Like FINRA

Demis Hassabis wants a new referee for AI, and he's borrowing the playbook from Wall Street. In a framework proposal published this week, the Google DeepMind CEO calls for a US standards body modeled on FINRA, the financial regulator, to evaluate frontier AI models. It would start with voluntary benchmarks funded by industry, then move toward mandatory rules as the technology matures. Startups and academic researchers would be exempt, a carve-out meant to head off accusations that big labs are using regulation to box out smaller competitors.

The proposal lands as Hassabis repeats a claim he first made in April: artificial general intelligence could arrive within a few years, with an impact ten times larger than the Industrial Revolution and ten times faster. In May, he described humanity as already standing "in the foothills of the singularity," a line that drew plenty of pushback. His pitch this time is less about prediction and more about mechanics, how to build oversight before the systems outrun the people building them. He also floats the idea that the standards body could coordinate a slowdown in development if the situation calls for it, an option Anthropic has reportedly weighed as well.