DeepMind CEO Demis Hassabis is pushing a plan for a self-regulatory organization to vet artificial intelligence models before they hit the market, borrowing a structure long used on Wall Street. The proposal, which he has been lobbying Washington officials to consider, would start with voluntary pre-release testing but could eventually become mandatory.
A Wall Street blueprint for AI
Hassabis wants to replicate the Financial Industry Regulatory Authority, or FINRA, for the AI sector. FINRA is a private, self-regulatory body that oversees broker-dealers under the Securities and Exchange Commission's watch. It writes rules, conducts exams, and fines firms that break them — all without being a government agency.
His pitch to lawmakers and regulators is that the AI industry needs a similar model: an independent organization that can set testing standards, review model safety, and enforce compliance, but with more agility than a federal regulator. The idea is to avoid a patchwork of state laws or a slow-moving federal bureaucracy.
Voluntary now, mandatory later
Under the plan, developers would agree to submit their most powerful models for pre-release testing. The tests would check for things like bias, misinformation risks, and potential for misuse. At first, it would be voluntary — companies could choose to participate. But Hassabis envisions that as the technology matures, the testing requirement would become legally binding, similar to how FINRA's rules are enforceable.
The timeline for that shift isn't clear. Hassabis has not said when he thinks the voluntary phase should end or what would trigger the switch. What he does argue is that waiting for a crisis before acting would be a mistake.
The Washington push
Hassabis has been meeting with officials in Washington to sell the concept. The lobbying effort is still in its early stages, and it's not yet clear which agencies or congressional committees would take the lead. The proposal would likely require legislation to give a self-regulatory body legal authority to mandate testing and impose penalties.
So far, there's no public sign of a bill. But Hassabis's backing carries weight — DeepMind is one of the world's leading AI labs, and its CEO has been an influential voice in the debate over AI safety. The challenge is that the industry is divided on how much regulation is needed, and some companies resist any form of mandatory oversight.
Hassabis's model would put the industry in charge of its own policing, but with a backstop of government enforcement. That's exactly how FINRA works. Whether Congress wants to hand that power to a private AI body is the open question.
