Sam Altman spent the week in Washington making a careful distinction. He is asking lawmakers to spend more money testing artificial intelligence, and asking them not to require AI companies to win government approval before releasing a model.
The first is a request for resources. The second is a request for a particular kind of regulation not to exist. OpenAI wants both, and the line between them is the whole argument.
Altman is urging Congress to reject proposals that would force developers to obtain federal sign-off ahead of a public launch, according to people familiar with his message. In place of an approval gate, he wants more funding for AI testing at the US Department of Commerce, and he wants the government to add scientists with expertise in cybersecurity, biological weapons, and national security to that effort. The pitch is for capacity to evaluate models rather than authority to block them.
The timing is not incidental. President Donald Trump signed an executive order asking AI companies to provide their models to the government for testing for up to 30 days before full release. The process is voluntary. Altman’s Washington visit is, in effect, an argument to keep it that way, to fund the evaluation while resisting any move that would convert a voluntary review into a mandatory licence.
He found a receptive room, at least in one office. Altman met members of Congress on Wednesday, including House Speaker Mike Johnson, who told CNBC he had a* “very good, productive meeting”* and described discussions about what a “light touch” regulatory framework should look like to “prevent some of the harms that could come from it.” The phrase “light touch” is doing the same work for Johnson that the testing-versus-approval distinction is doing for Altman.
The position is consistent with where OpenAI has landed before. The company has supported the idea of evaluation and disclosure while resisting pre-clearance regimes, the version of regulation that would put a government office between a finished model and its users.
It is also the more conciliatory register from a chief executive who has, at other moments, downplayed the technology’s risks and argued an AI jobs apocalypse is unlikely. Funding Commerce to test models keeps the expertise inside government without handing it a veto.
The body Altman wants funded is the AI testing effort housed at the Commerce Department, which OpenAI argues should grow and add specialists in cybersecurity, biological weapons, and national security, the domains where a frontier model could plausibly cause the most damage.
The implicit bargain is that government should be equipped to find problems in advance, but should do so through testing it pays for rather than through a licence it issues. Where the line falls between the two is precisely what Congress would decide if it legislated.
Altman is not the only voice in the room. The visit comes amid a broader push by AI companies to shape how Washington translates Trump’s executive order into durable policy, and the “light touch” framing Johnson used is the one the industry has been encouraging. Whether that framing survives contact with members of Congress who favour harder rules is the part the week did not resolve.
None of this is settled. Altman is lobbying against proposals, not against enacted law, and the executive order he is responding to remains voluntary on its face. What the week established is the shape of the fight ahead. The administration has asked for a look before release. Altman is willing to give the look, and is in Washington to keep it from becoming a gate.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.