Big AI Labs Court Krishnan to Lead Their Own Regulator

Two days after their own CEOs told the UN Security Council that AI risk needs international cooperation, Google, OpenAI and Anthropic are reportedly building something narrower: a private standards body they would run themselves. The Information reported Wednesday that the three labs are lining up a self-regulator, tentatively called the Frontier AI Standards Agency, and have approached Sriram Krishnan, the Trump administration's outgoing senior AI policy adviser, to be its CEO. Krishnan spent his eighteen months in the White House arguing against exactly this kind of body existing under government control. Now the same three companies want him to run one of their own.

Table of contents

• What's being proposed
• The CEO pick is the whole story
• This isn't the labs' first attempt
• The structural question nobody has answered
• The timing against the UN session
• What would actually make this real
• Frequently asked questions

What's being proposed

According to AI Weekly's summary of The Information's reporting, the agency would be modeled on FINRA, the body Wall Street uses to police its own brokers without a government regulator doing it directly. The pillars under discussion, per multiple outlets tracking the story, include shared technical evaluations, pre-release audits, independent testing frameworks and standardized safety protocols across the three companies' models.

OpenAI's Chief Global Affairs Officer Chris Lehane confirmed on September 15, in a Washington briefing reported by AI Weekly, that the three labs had already been coordinating on safety protocols for several weeks before the story broke. Stocktwits' roundup adds that OpenAI has publicly said it intends to proceed "with or without government support." A target launch of late 2026 or early 2027 has been floated, though nothing has been finalized: Krishnan has not accepted the CEO role, and neither Google nor Anthropic has publicly confirmed its own participation on the record.

The CEO pick is the whole story

Krishnan served as the White House's senior AI policy adviser from January 2025 to June 2026, a period Time credited him with shaping into a $500 billion national AI push. On his way out, byteiota's coverage notes, his explicit position was that "there will not be an FDA for AI," and that a centralized regulator requiring "a team of lawyers before you can get a model out" would put "sand in the gears" of AI progress.

Byteiota's framing of the hire is blunt, and worth repeating because it captures the tension precisely: the three biggest AI labs want the most prominent recent opponent of AI regulation to run their own regulator. That is not necessarily a contradiction if the goal really is voluntary industry coordination rather than government-style enforcement, but it does mean the person most likely to lead this body has already staked out a public position against the kind of binding rules a skeptic would want to see it eventually enforce.

This isn't the labs' first attempt

Byteiota's reporting points to a precedent that undercuts some of the novelty here. Google, Microsoft, OpenAI and Anthropic formed the Frontier Model Forum in 2023, with a $10 million safety fund and a mandate to advance responsible AI development. In byteiota's assessment, it has published best practices and has never stopped a single release, required a single evaluation, or enforced a single standard.

The pitch for the new agency is that it would be structurally different: more formal, with auditor certifications and pre-release requirements rather than voluntary best-practice papers. Whether that difference holds up in practice is untested, since the Forum was also announced with confident language about raising the safety bar industry-wide.

The structural question nobody has answered

The same structural conflict that limited the Forum applies here in a more direct form. Byteiota states it plainly: the companies being evaluated would fund and staff the body doing the evaluating. A self-regulator paid for by the three largest labs in the field has an obvious incentive problem, whatever its formal charter says about independence.

Byteiota also raises a second-order effect worth watching: if the agency standardizes pre-release evaluations, access windows before launch could get longer, meaning release timelines could stretch out, and model capability at launch could be constrained to whatever passed the safety threshold at evaluation time, rather than the full capability the underlying model supports. That is a plausible outcome of any serious pre-release audit process, not unique to this specific proposal, and it cuts against the frequent industry argument that safety review slows a company down without saying anything about whether the slowdown is worth it.

The timing against the UN session

The AI Daily Digest's framing of the timing is the sharpest available: two days after the three labs' leadership told the UN Security Council that AI risk needs international cooperation, detailed in our report on the UN Security Council AI briefing, the same companies were reported to be building their own private standards body instead of waiting on that international process. Anthropic's Dario Amodei has separately called for international measures on frontier AI risk, including shared verification systems between countries. A voluntary, industry-funded, US-only body is a narrower thing than that.

Neither framing is necessarily wrong. A private industry standard could be a genuine first step that a slower international process eventually builds on, or it could be a way to demonstrate self-governance persuasively enough that binding international rules never get built. Nothing in the reporting reviewed here settles which one this becomes, and the companies involved have not addressed the sequencing directly.

What would actually make this real

A few concrete developments would separate a working regulator from an announcement. Whether Krishnan accepts the role, and on what terms, is the first signal. A published founding charter with actual enforcement mechanisms, rather than best-practice language, would be the second. Whether Google and Anthropic confirm participation on their own terms, rather than through leaks and third-party reporting, is a third. And whether any non-founding lab, or any government body like NIST, gets a seat in the testing pipeline would say something about whether "independent" describes the agency's structure or only its name.

Frequently asked questions