The FINRA for AI: Google, OpenAI, and Anthropic Are Building Their Own Safety Watchdog

The three rival AI labs are reportedly finalizing plans for an industry-run Standards Authority for Frontier AI, aiming to launch by early 2027 — with former White House AI advisers in the frame to run it.

Published September 29, 2026 · By Daniel Rao

Google DeepMind CEO Demis Hassabis delivering a lecture at EPFL in May 2023. His July proposal for a FINRA-style AI standards body helped set the SAFA talks in motion. Photo: Alain Herzog / Mediathèque EPFL, via Wikimedia Commons (CC BY-SA 4.0).

The fiercest rivals in artificial intelligence are, reportedly, about to become co-founders of its first industry-run watchdog. Google, OpenAI, and Anthropic are finalizing plans to establish an independent body — provisionally called the Standards Authority for Frontier AI, or SAFA — that would set common benchmarks and testing practices for frontier AI models. According to reporting from The Information, corroborated this week by multiple outlets, the body could launch by late 2026 or early 2027.

The plan would move ahead without the federal supervision that was originally envisioned. Talks began around a public-private partnership under federal oversight, but that idea stalled after a White House draft executive order failed to secure enough backing inside the Trump administration, according to The Information's reporting. Facing that impasse, the three labs pivoted to self-regulation: an industry-funded, industry-run organization operating independently of the government.

An OpenAI spokesperson has confirmed the company is in active discussions with Google and Anthropic about coordinated safety frameworks, according to CNBC reporting cited by ET CIO. That is about as close to an official acknowledgement as this story has received. None of the three companies has announced SAFA itself — everything below is reported, not confirmed — and the most basic questions about the body remain unanswered. But the fact that the talks have advanced this far, with leadership candidates already being approached, suggests the industry's long-theoretical conversation about self-regulation is about to become very concrete.

Where the idea came from

The intellectual origin of SAFA is a proposal Google DeepMind CEO Demis Hassabis published in mid-July: a U.S.-backed standards body for AI structured as a public-private partnership or self-regulatory organization, modeled explicitly on Wall Street's Financial Industry Regulatory Authority (FINRA). FINRA is the private, not-for-profit organization authorized by Congress to police broker-dealers — the analogy being that AI's biggest labs could submit to a shared, technically expert overseer rather than wait for Washington to build one.

The proposal gained momentum this month after Anthropic CEO Dario Amodei published an essay calling for an industry-wide slowdown in the release of frontier models, arguing that safety mechanisms need time to catch up with capabilities. And last week, the stakes went global: OpenAI's Sam Altman and Amodei told the United Nations Security Council on September 23 that the industry needs stronger oversight, a remarkable moment of public alignment from two companies whose commercial interests otherwise diverge sharply.

Meanwhile, Washington's own evaluation infrastructure is widely seen as inadequate. The U.S. Center for AI Standards and Innovation already handles federal model evaluation, but people familiar with the SAFA discussions told The Information that the agency lacks the technical talent and funding to thoroughly inspect increasingly complex frontier systems. The labs, in other words, appear to have concluded that if a credible evaluator is going to exist, they will have to build it themselves.

Who could run it

The leadership outreach is strikingly bipartisan. The three companies have reportedly approached Sriram Krishnan, the Trump administration's AI policy adviser and a former venture capitalist, and Arati Prabhakar, who directed the White House Office of Science and Technology Policy under the Biden administration, about the chief executive role. Reaching across administrations to two people who have sat on opposite sides of the same desk is a deliberate signal: this body wants to be seen as serving American interests, not the labs' interests.

Former Secretary of State Condoleezza Rice is reportedly under consideration for the board chairmanship, and venture capitalist David Friedberg has also been named in leadership discussions, according to CNBC TV18's reporting. Potential scientific consultants include Beth Barnes, founder and CEO of the AI evaluation nonprofit METR, and Paul Christiano, an adviser to the Center for AI Standards and Innovation — names that carry real weight in the small, technical world of AI safety evaluation.

Related: Trump, Johnson and Tech CEOs to Meet on AI — our report on today's White House AI summit, the backdrop against which the SAFA talks are unfolding.

Why the timing is impossible to ignore

SAFA's emergence coincides with a decisive week for AI policy in Washington. Today — Tuesday, September 29 — President Trump is meeting House Speaker Mike Johnson and top tech executives to discuss AI policy, a session the president has described as "very important." Speaking to reporters, Trump framed the administration's posture in bluntly competitive terms: "I'm for let's go and let's win," he said. "Because I think if you… open it up to China, you give up the lead." That is the same logic that reportedly sank the federal-supervision version of the standards body: the White House sees speed as the priority, and formal slowdown mechanisms as a strategic liability.

That tension is the engine of the SAFA story. The labs want a credible way to demonstrate their models are safe — partly to reassure the public, partly to preempt the kind of regulation they fear more. The administration wants American labs to win the race with China and views restraint as a handicap. An industry-run body threads that needle: it lets the labs claim independent oversight without asking Washington to bless a slowdown, and it lets the White House point to self-regulation instead of building new federal machinery.

The questions nobody has answered — our analysis

Everything above is reported. What follows is analysis, and it is worth being direct about the gaps. The SAFA talks have not resolved — publicly, at least — the three questions that will determine whether the body matters.

First, enforcement. FINRA works because it has teeth: it can fine firms, suspend brokers, and bar bad actors from the industry. What powers would SAFA have? According to the reporting, its standards and enforcement powers have not been announced. A body that publishes benchmarks but cannot compel compliance is an advisory panel with a fancier name. The labs are still debating even the basic scope of the job: whether SAFA should conduct model assessments in-house, or merely set the standards and qualifications for third-party evaluators and auditors who do the actual testing.

Second, governance and capture. A watchdog funded by the companies it watches faces an obvious conflict of interest — the same structural criticism leveled at credit-rating agencies before 2008 and at FINRA itself by Wall Street's critics. The bipartisan leadership outreach is clearly designed to answer that criticism in advance. But until there is a public charter, a funding formula, and a governance structure that gives genuine power to people who are not the labs, skepticism is the rational default.

Third, the binding question: what happens if SAFA evaluates a model and finds it unsafe? Would the labs commit in advance to delaying or withholding deployment — the kind of commitment that could cost billions in a race they are desperate to win? Or would an adverse finding be published as guidance that each company is free to weigh against its own commercial judgment? Nothing reported so far answers this. And without an answer, SAFA risks becoming a very expensive rubber stamp.

None of this means the initiative is hollow. Pre-deployment testing standards, incident-reporting norms, and auditor qualifications are genuinely useful public goods, and the U.S. government demonstrably lacks the capacity to produce them right now. If SAFA can produce credible evaluation infrastructure that Congress, insurers, and enterprise buyers can lean on, it will have done something real — even without FINRA-style teeth. The test will be whether the first evaluation that goes against a lab's commercial interests is honored or quietly buried. Watch for the charter, the funding, and the first hard call. Everything else is branding.

Sources: The Information (via TechRepublic, The Times of India, ET CIO, AI Tech Daily, The Prompt Insider, TechGig); CNBC (via ET CIO); CNBC TV18 (via AI Tech Daily); The Verge (via AI Tech Daily); USA Today (Trump remarks on the September 29 AI meeting).