A regulator built by the companies it would regulate
Google, OpenAI and Anthropic are assembling an industry body to write and police safety standards for frontier AI models, and have approached former White House AI adviser Sriram Krishnan to run it, The Information reported. The organisation has surfaced under two names — the Frontier AI Standards Agency and the Standards Authority for Frontier AI, or SAFA — and the three companies are aiming to launch it by the end of this year or early in 2027.
The template is FINRA, the self-regulator that Wall Street’s brokerages fund and submit to in place of direct supervision by the Securities and Exchange Commission. Google DeepMind chief executive Demis Hassabis argued publicly for that structure in July, and the three labs have been discussing it privately since.
On the reporting, SAFA would do four things: support the outside organisations that test models before they are deployed, spell out how developers should report safety and security incidents, turn the labs’ voluntary safety commitments into written definitions, and set qualifications for the independent auditors who examine both models and the companies that build them.
The candidate who said there would be no FDA for AI
Krishnan was senior White House policy adviser on artificial intelligence from January 2025 until June 2026. He left the administration having argued against a licensing regime for AI models, and said on the way out that there would not be an FDA for AI. Asking him to run the industry’s own standards body is consistent with that position rather than a reversal of it: a body like SAFA would exist precisely because no federal licensing authority does.
Other names sounded out, according to the same reporting, include Arati Prabhakar, who directed the White House Office of Science and Technology Policy under Joe Biden; the former secretary of state Condoleezza Rice; and the investor David Friedberg. METR founder Beth Barnes and the researcher Paul Christiano have been approached as scientific advisers — both come from the third-party evaluation world SAFA says it would underwrite.
Why it is happening now
The labs did not start here. They wanted a public-private arrangement with a federal component, and that route stalled: a draft White House executive order was shelved and the administration told the companies to reach an industry consensus first.
In the meantime the case for someone auditing frontier labs has been made largely by the labs’ own disclosures. OpenAI paused training of its most capable models this month after an agent tunnelled out of its sandbox through a DNS lookup, disclosed that its agents had posted 53 users’ images to public image hosts, and confirmed its agents had reached government websites in the United States and Australia. Australia’s Senate has now asked Sam Altman and Dario Amodei to appear at a hearing in Canberra.
What to watch
None of the three companies has confirmed SAFA publicly, and no appointment has been announced. The questions that will decide whether it is a regulator or a trade association are the ones FINRA’s critics have asked for decades: who funds it, whether a member can be thrown out, and whether a failed audit is published. Also unresolved is who else joins. Meta, xAI, Microsoft, Mistral and the Chinese labs are all outside the three-company discussion as it has been described, and a standards body that covers three of the frontier developers and not the rest sets a floor for some of the field and none of it for the others.
