Demis Hassabis, CEO of Google DeepMind, has proposed creating a new US regulatory body to oversee advanced AI development, modeled after the Financial Industry Regulatory Authority (FINRA). The proposal reflects Hassabis's view that uncertainty about AI's trajectory demands immediate guardrails.
The framework centers on a standards body that would develop evaluation protocols for frontier AI models and retain authority to coordinate industry-wide development slowdowns if risks escalate. Startups and academic research would receive exemptions from these requirements, preserving innovation in smaller-scale projects.
Hassabis frames the approach as "cautious optimism" rather than either accelerationism or heavy-handed restriction. His statement that "nobody in the world knows what happens next" underscores the core tension: advanced AI capabilities are advancing rapidly while their long-term consequences remain uncertain. Rather than waiting for definitive evidence of harm, Hassabis argues for proactive measurement and coordination now.
The FINRA model offers a specific precedent. That organization operates as a self-regulatory authority within financial markets, setting standards and enforcement mechanisms while remaining distinct from government regulators. Applying this hybrid approach to AI would create an industry-led body with teeth, capable of slowing development without requiring Congressional action.
The proposal sits at the center of ongoing AI governance debates. It avoids both libertarian opposition to regulation and calls for immediate AI development bans. Yet questions persist about enforcement power, international coordination, and whether a FINRA-style body could actually enforce a slowdown when economic incentives favor speed.
Hassabis's framing reflects DeepMind's position as both a leading AI developer and a company concerned about risks. This dual stance shapes his regulatory vision: structures that can manage risks without strangling progress. The proposal suggests major AI labs increasingly view some form of coordination and oversight as preferable to either