DeepMind CEO calls for an independent body to test frontier AI models

Google DeepMind CEO Demis Hassabis has proposed establishing an independent body to oversee the most advanced AI models. Hassabis envisions the organization operating in a manner similar to self-regulatory bodies that already exist within the finance industry.

The example he points to is a self-regulatory organization in the United States that oversees securities brokerage firms. Although not a direct government agency, it sets rules that firms in the sector must follow and carries out compliance reviews.

Under Hassabis's proposal, a similar structure could be built for AI. The body would test so-called "frontier" models — those operating at the cutting edge of current capability — before public release, evaluating them against agreed safety standards.

The idea is being framed as a middle path between AI companies regulating themselves entirely and governments intervening directly. Hassabis argues that regulation controlled entirely by government agencies may struggle to keep pace with a fast-moving technology.

Critics counter that industry self-regulation carries an inherent conflict-of-interest risk. If a body's funding comes largely from the very companies it oversees, its independence could reasonably be questioned.

Different regulatory approaches to AI already coexist globally. The European Union has built a comprehensive legal framework, regulation in the United States remains largely fragmented and state-by-state, while China applies its own centralized oversight model.

Hassabis argues that his proposed independent body could serve as a bridge across these differing regulatory approaches, letting companies rally around a shared standard rather than navigating separate rules in every jurisdiction.

For the proposal to work in practice, major AI companies would need to accept the authority of such a body voluntarily. Without that buy-in, the organization would have little practical enforcement power.

Some industry observers note that similar proposals have surfaced before without ever taking concrete shape. The intensity of competition among AI companies can make it difficult to agree on a shared oversight framework.

Hassabis's call underscores that the debate over AI safety isn't confined to governments and civil society groups — it's also playing out among the industry's biggest players. Whether the proposal turns into an actual institution remains an open question.