Chris Lehane, OpenAI’s chief global affairs officer, stood before reporters on September 15 and said something the company had never said before. OpenAI, he announced, supports the key provisions of a bipartisan House bill that would require leading AI labs to let outside evaluators inside their model development.
The bill is called the FRONTIER Act, and it would build the first federal framework for AI safety in the United States. It was introduced by Representative Jay Obernolte, a Republican, and Representative Lori Trahan, a Democrat. Its central mechanism is an “independent verification organization,” or IVO, that would be granted access to inspect how frontier models are built.
Lehane said he had told one of the bill’s sponsors exactly where OpenAI stands during a meeting on Capitol Hill on Monday. The statement was not a vague gesture of support. It was an endorsement of a specific, federally mandated, third-party evaluation requirement, and it was the first time OpenAI had gone that far.
The move fits a pattern that has accelerated over the past two weeks. When California Governor Gavin Newsom signed the state’s AI safety bill last week, OpenAI added its support the same day. On Saturday, Chief Executive Sam Altman promised to voluntarily embed third-party evaluators inside the company, hours after a similar pledge from Dario Amodei of Anthropic.
The sequence suggests a strategy. By supporting oversight publicly and early, OpenAI positions itself as a partner in writing the rules rather than a target of them. A company that helps draft the framework is more likely to live comfortably inside it than one that fights it until the end.
A staffer on Trahan’s office described OpenAI’s endorsement as a very strong signal. But signals, the staffer implied, are not laws. The next few weeks will show whether the commitment survives contact with the legislative process, where details on access, liability, and the independence of the evaluators still have to be settled.
The FRONTIER Act addresses a real gap. The United States has no federal regime governing frontier AI, and the current approach has been a patchwork of voluntary commitments and state-level rules. A federal IVO requirement would change the balance of power between labs and the government for the first time.
Critics inside the industry have argued that mandatory inspection could slow development and leak trade secrets. OpenAI’s endorsement pushes back on that view, arguing in effect that a lab can open its doors to evaluators and still move fast enough to compete.
The political timing is delicate. OpenAI’s support comes as the White House has framed AI leadership in national-security terms, and as some in Washington question whether any slowdown serves the country’s interests. A company endorsing federal oversight while the president dismisses doom warnings as a con is holding two very different conversations at once.
What OpenAI gets in return is legitimacy. The company has faced skepticism about whether its safety commitments are marketing or substance. Attaching its name to a specific piece of legislation, with specific enforcement teeth, is a way to answer that skepticism in public.
Analysts said the endorsement also has a competitive dimension. If federal rules arrive, the labs that shaped them will bear them more easily than those that did not. Being the first to endorse the framework is a way to define the floor that rivals will have to meet.
The risk is that the bill changes in ways OpenAI does not like, and that its early support becomes a constraint. Endorsing a mandate is easy; living under one written by a divided Congress is harder. The company has bet that influence now is worth the risk of constraint later.
What an IVO would actually do is still being defined, but the outline is clear. An independent organization would be granted access to inspect model development before release, looking for the capabilities and failure modes that a company’s own tests might miss. The fights over how much access, and what happens when an evaluator flags something, are exactly the fights the bill will now attract.
The California sequence gives the endorsement context. When Newsom signed the state’s safety bill, OpenAI’s support came the same day, a pattern of joining regulation after it has passed rather than opposing it in advance. Altman’s Saturday pledge to embed third-party evaluators voluntarily was the next step in the same direction, and the federal endorsement extends it.
The bipartisan pairing of Obernolte and Trahan is the reason the bill has a chance. AI safety has become a rare issue where members of both parties see an opening, and a company endorsing that coalition is positioning itself for the moment when a divided Congress is still capable of writing a bill it can live with.
For now, the endorsement is a marker of where the industry’s center is moving. The debate is no longer whether AI should be inspected, but by whom, under what terms, and how much of the process the labs themselves will control.
