Reported 15 Sept 2026 Governance

OpenAI told Congress it could accept outside inspectors written into federal law, and asked Britain to legislate on the same day.

Politico reported on September 15 that OpenAI's lobbyists told Capitol Hill the company can support the part of the FRONTIER Act that would require the largest AI developers to bring in independent verification organizations, the outside bodies that would check model safety claims; one person present quoted OpenAI as saying it made clear that we can support that. The same day OpenAI's European policy lead, Tom Duff Gordon, publicly urged the United Kingdom to pass binding rules on frontier developers and to take advantage of a political window that is clearly opening.

The caveat is that both are accounts of what OpenAI said in rooms and to reporters: no OpenAI publication names the bill, and the FRONTIER Act would also set one national standard in place of state laws, which is the thing the industry has wanted for two years.

Until now

The labs had promised outside evaluators voluntarily, in essays and interviews, and the only binding version of the idea was a state statute in California.

Now

Treat mandatory third-party verification as a provision with the largest lab's stated support inside a live federal bill, and treat the preemption of state law as the price that support is being offered for.

For two weeks the pacing argument has been made in essays and answered in share prices. On September 15 it turned into a legislative position. According to Politico, OpenAI's lobbyists told people on Capitol Hill that the company can support the part of the FRONTIER Act that would require the largest developers to use independent verification organizations, the outside bodies that check a lab's safety claims the way an auditor checks accounts. One person in the room quoted OpenAI as saying it made clear that we can support that. This is the same idea Dario Amodei committed Anthropic to on September 12, that Sam Altman agreed to hours later, that Satya Nadella endorsed on the 13th and that California turned into a voluntary legal advantage on the 9th. The difference is that this version would be mandatory and federal, and the company being asked to accept it is the one with the most models to inspect.

Then read the rest of the bill. H.R. 9925 does not only create the inspectors. It sets a uniform national standard for transparency, auditing and catastrophic-risk reporting, which means one federal rule in place of California's, New York's and everyone else's, and that is the thing the industry has been asking for since the first state law passed. Support for the inspection provision and support for the preemption provision arrive in the same envelope, and only one of them costs OpenAI anything. On the same day, in London, OpenAI's European policy lead asked the British government to legislate on frontier developers and to use a political window he described as clearly opening. None of this is published by OpenAI; it is reported from rooms and from remarks, which is why this card runs as Reported with a verification event on October 31. Treat mandatory third-party verification as a provision with the largest lab's stated support inside a live federal bill, and treat the preemption of state law as the price that support is being offered for.

How it got here

Jul 23
Jay Obernolte and Lori Trahan introduce H.R. 9925, the FRONTIER Act, with four other members. It is referred to two committees and nothing further happens to it.
Sep 09
California creates the first system of independent AI auditors and makes using one a legal advantage. Carded in Edition 017.
Sep 12
Anthropic commits to embedded outside evaluators and OpenAI says the same hours later. Carded in Edition 020.
Sep 14
The President calls AI safety concerns a hoax. Microsoft publishes a code of conduct that covers only the models it builds. Recorded and carded in Edition 022.
Sep 15
OpenAI tells Capitol Hill it can support the bill's independent verification organizations, and its European policy lead asks the United Kingdom to legislate. The Pro-Human Assembly meets in Washington with Sanders, Bannon and four members of Congress.
Who has to react
Obernolte and Trahan, whose bill went from stalled to endorsed by the largest company it would bind; the California legislators and the new state commission, whose statute this bill would supersede; Anthropic and Google, who have been in the same rooms and have not said the same thing on the record; the White House, which called the safety argument a hoax the day before; the auditors and evaluation labs who would become the independent verification organizations, and who have no accreditation regime yet; and every state attorney general holding an open AI matter that a national standard would close.

The evidence

The endorsement is reported by Politico from meetings on Capitol Hill and corroborated by the Washington Examiner; OpenAI has published nothing naming the FRONTIER Act, and its global affairs page carries no post dated September 15 or 16. Verification event, October 31, 2026: OpenAI publishing a position that names H.R. 9925 or independent verification organizations, a committee witness list, hearing record or markup document recording its support, or a third-quarter federal lobbying disclosure naming the bill. Absent one of those, this stays Reported and we will say so.

ObservedTuesday September 15, 2026. Politico's Brendan Bordelon reported that OpenAI is backing a provision of the FRONTIER Act, the bipartisan House bill, that would require top AI companies to embed outside evaluators to check model safety, with one person present quoting OpenAI as having specifically talked about how we think about IVO, meaning independent verification organizations, and made clear that we can support that.

The whole observation, as recorded

The Washington Examiner carried the same story, describing OpenAI lobbyists backing the regulatory provisions of a bipartisan bill that creates a framework pathway forward for the regulation of frontier artificial intelligence development. The bill is H.R. 9925, introduced July 23, 2026 by Jay Obernolte of California with Lori Trahan of Massachusetts and four others, referred to Energy and Commerce and to Science, Space and Technology, with no further action recorded on the House calendar as of this morning. Its published design: tiered requirements scaled to the size of a frontier developer, covering model cards, risk-management frameworks, independent audits, incident reporting and ongoing assessment, establishing a uniform national standard for transparency, auditing and reporting of catastrophic risk in place of varying state rules. Separately on September 15, TechCrunch reported that OpenAI, Anthropic and Google have been in talks on safety standards for weeks; Bloomberg reported OpenAI saying it has worked with the other two without needing an antitrust exemption; and The Information reported working groups below chief executive level discussing an industry-led standards body. Also on September 15, OpenAI's European policy lead Tom Duff Gordon publicly called on the United Kingdom to pass binding rules confined to the handful of companies developing the most powerful AI systems, and urged the government to take advantage of this political window which is clearly opening up. The surrounding week on the record: Amodei's pacing essay of September 12 and Altman's agreement, carded in Edition 020; Nadella's backing of embedded evaluators on September 13 and Microsoft's code of conduct on September 14, carded in Editions 021 and 022; California's independent auditor system created September 9, carded in Edition 017; and the President's post on September 14 calling AI safety concerns a hoax, recorded in Edition 022. OpenAI's own global affairs page carries posts dated September 9 and September 10 and nothing dated September 15 or 16.

Established

Politico reported on September 15, 2026 that OpenAI told people on Capitol Hill it can support the FRONTIER Act's independent verification organization provision, with a direct quotation from a person present.
The Washington Examiner independently reported OpenAI lobbyists backing the bill's regulatory provisions on the same day.
OpenAI's European policy lead publicly urged the United Kingdom to pass binding rules on frontier developers on the same day.
H.R. 9925 exists, was introduced on July 23, 2026 by Obernolte and Trahan, and has had no action beyond referral to two committees.

Not established

That OpenAI has endorsed the FRONTIER Act. The reporting describes support for one provision, relayed by attendees, and OpenAI has published nothing naming the bill.
That the bill will move. It has sat in two committees since July with no markup scheduled, and the President has spent the week calling the underlying concern a hoax.
That mandatory inspection is coming. No accreditation regime for independent verification organizations exists at federal level, and the bill's text would still have to survive committee.
That this is a concession. The same bill would replace state AI laws with one federal standard, which is what the industry has lobbied for since 2024.

Why it weighs this much

Signal strength

CostLow

CommitmentMedium

ReversibilityEasy

PersistenceMonths

Cost is low. Saying you can support a provision of a bill that has had no markup, no floor time and no Senate companion costs nothing today, and the bill would hand the speaker a national standard worth more than the inspection it accepts. Commitment is medium rather than low only because the statement was made to legislators who will remember it, and because a company that says this in September cannot easily lobby against it in March without that being visible in a filing. Reversibility is easy: positions relayed by a reporter can be clarified away, and nothing has been signed, filed or published. Persistence is months, because the bill either moves in this Congress or dies with it. The bars are deliberately modest, and they are the honest answer to a story that reads much bigger than it is.

What to watch

strengthens The FRONTIER Act is marked up in the House Energy and Commerce Committee, or OpenAI publishes a position naming H.R. 9925 or independent verification organizations, by March 31, 2027.
weakens No markup is held and no OpenAI publication, hearing record or federal lobbying disclosure names H.R. 9925 by March 31, 2027, or OpenAI supports a preemption measure carrying no independent verification requirement.

The words

Independent verification organization: An outside body accredited to check a company's safety claims about its own models, in the way an auditor checks accounts. The FRONTIER Act would make using one a legal requirement for the largest developers.
Preemption: A federal law overriding state laws on the same subject. For a company facing different rules in California, New York and Colorado, one national standard is worth a great deal, whatever it contains.
Markup: The committee meeting where a bill is amended and voted on. Until a bill is marked up, support for it costs nothing.
Lobbying disclosure: A quarterly filing naming the bills a company paid lobbyists to work on. It is the cheapest way to check whether stated support was real.

Your call

If the reporting holds, which explanation do you believe?

A call is final. Other people's calls show after yours.