The Big AI Labs Want to Regulate Themselves

While the public argument over AI safety plays out on social media, a quieter conversation has been running in conference rooms since July. Representatives from Google, OpenAI and Anthropic have met repeatedly to discuss building an AI standards body, one the industry would run itself, CNN reported this week.

The idea did not start with Anthropic's weekend essay, though the two fit together neatly. It traces to a July piece by Demis Hassabis, the Google DeepMind founder, who proposed a United States standards body modelled on the Financial Industry Regulatory Authority, the private organisation that polices Wall Street brokers. Hassabis argued such a body should test advanced models before they ship, run as a public-private partnership, overseen by government but paid for by the industry and staffed by independent technical experts and open-source representatives, according to PYMNTS.

Agreement, up to a point

Sam Altman has said he favours some kind of testing and auditing organisation. He also thinks the labs will have to build it on their own, without waiting for Washington to bless it, Quartz reported. That reading looks sharper by the day. The president spent the weekend rejecting the whole premise that AI needs guardrails, which makes a government-anchored regulator a hard sell for now.

Self-regulation carries an obvious tension. A body funded by the companies it inspects has to prove it can say no to them, and the FINRA comparison cuts both ways, since Wall Street's self-regulator has drawn plenty of criticism for being too soft on its members. Supporters counter that a private body can move at the speed the technology actually moves, which no legislature can.

For now this is talk, not an institution. There is no charter, no leadership, no first model queued up for review. But three rivals who guard their research closely are sitting at the same table trying to write the rules they will all live under. That alone is worth watching.