Google DeepMind CEO Demis Hassabis made a notable proposal on July 14, calling for the creation of a US-led independent body to test the most powerful AI models before they go live for American users. The proposed structure is explicitly modeled on FINRA, the Financial Industry Regulatory Authority that oversees broker-dealers in the US financial system.

The proposal is significant coming from the head of one of the world’s most advanced AI labs. It signals that at least some of the people building frontier AI want formal oversight built into the system, not just voluntary commitments.

What the Proposed Body Would Do

Hassabis envisions a federally overseen public-private partnership, funded by the AI industry itself. The board would include technical experts, open-source AI representatives, and government officials.

The body’s job would be to evaluate the most capable AI models for risks across three specific threat categories: national security, cybersecurity, and biological threats. Both open-source and closed-source models would fall under its scope.

The proposed process works like this: AI labs would voluntarily share their frontier models with the body for review up to 30 days before public release. Over time, that voluntary participation would become mandatory. Compliance would eventually become a prerequisite for any model to go live for users in the United States.

Hassabis described wanting this body in place before the end of 2026.

Why Now

This is not the first time Hassabis has raised the topic of frontier model governance. But the July 14 proposal is more concrete than previous statements. It names a specific structural model (FINRA), a specific timeline (before year end), and a specific scope (the most advanced AI systems, not all AI tools).

The timing matters. The US Congress has been debating AI regulation for months, with competing proposals on whether to let states set their own rules or establish a federal standard. The AI industry has lobbied hard to prevent what many see as inconsistent state-level requirements. But so far, no unified framework has emerged.

Against that backdrop, Hassabis is essentially offering a third path: industry-funded, technically expert oversight that sits between self-regulation and full government control.

This Is Not Yet Policy

It is worth being clear about what this proposal is not. It is one CEO making a public argument. There is no legislation, no regulatory action, and no concrete timeline beyond Hassabis’ stated ambition.

The FINRA analogy is useful for understanding the intent but comes with caveats. FINRA regulates financial advisors, not the underlying financial products themselves. A true parallel for AI would be something closer to how the FDA tests drugs before they reach patients. The exact scope and authority of any future body remains completely open.

Other major AI labs, including OpenAI and Anthropic, have taken different positions on AI regulation recently. OpenAI has pushed for a federal framework that pre-empts inconsistent state rules. Anthropic has supported state-level AI safety laws. Hassabis’ proposal sits outside that debate, focused specifically on technical model evaluation rather than broader liability or transparency requirements.

What This Means for Business

For enterprise leaders evaluating AI tools and vendors, this kind of discussion matters even before it becomes law.

The direction of travel across the industry is toward more formal accountability for frontier models. Whether or not Hassabis’ specific proposal takes shape, the AI regulatory environment is becoming more complex. Businesses that are already building internal governance processes around AI are better positioned to adapt as requirements tighten.

A few practical considerations worth tracking:

Model selection will carry compliance implications. If testing standards do emerge, models that have passed a formal review process will carry different weight in procurement decisions than those that have not. Enterprise buyers will increasingly ask vendors about compliance posture, not just capability benchmarks.

Your AI vendor relationships matter. Businesses that have built AI workflows on top of a single frontier model face concentration risk if that model’s deployment or access terms change due to regulatory pressure. A diverse, governed approach to AI tool selection reduces that exposure.

Governance is becoming a differentiator, not a cost. The companies that treat AI governance as a serious operational function, rather than a compliance checkbox, are the ones that will move faster as clarity eventually arrives. The tools, processes, and internal expertise built now will compound in value as requirements get more specific.

Smaller teams have more to gain. Enterprise-scale AI labs have legal and policy teams working on this full time. For smaller and mid-sized businesses, the implication is that your AI partners need to take compliance seriously on your behalf. Choosing vendors who participate in voluntary safety evaluations now is a reasonable proxy for which ones will be prepared when requirements become mandatory.

---

Enterprise DNA works with businesses across industries to build AI capabilities that are practical, governed, and built for the long term. If you want to understand how emerging AI regulation might affect your business, our advisory team can help you think it through. Talk to us about your AI strategy.