Agentic market $10.8B and climbing  ·  editor@gaasnews.com
Sections
HomeWhat is GaaS?PlatformsPricingGlossaryOpinionAboutContact
HomePolicy & RegulationOpen weights lobbying
Policy & Regulation

OpenAI and Anthropic quietly lobbied Washington to restrict open models, NYT reports

The New York Times reports the two closed-model labs pressed regulators to treat model weights as protected intellectual property, while 25 companies from Nvidia to Mozilla signed a letter the labs sat out.

AJ
Andrew Jamerson
Founding Editor
Jul 26, 2026 · 5 min read
Twenty-five signatures on one side of the split, five unnamed sources on the other. // GaaS News

OpenAI and Anthropic have quietly lobbied regulators in Washington to restrict open-source AI models, The New York Times reported Saturday, citing five unnamed sources, even as OpenAI chief executive Sam Altman publicly says he supports open source. The reported push has drawn in Treasury Secretary Scott Bessent and White House science adviser Michael Kratsios, who have weighed in on treating American model weights as valuable intellectual property.

What the Times reports

Per the Times account, republished by GV Wire, the campaign is driven by the rise of two Chinese startups, Z.ai and Moonshot AI, which have released models that rival those from Anthropic and other American labs. Both US companies have claimed that Chinese firms built those models by improperly harvesting data from their AI systems, a practice Kratsios has called "large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology." Publicly, executives such as Altman have said they support open-source software; the Times frames the lobbying as running quietly underneath that public posture.

OpenAI is separately urging the administration to adopt policies requiring mandatory security evaluations of new AI models overseen by government agencies, per the report. The lobbying account rests on the Times' unnamed sourcing, and the paper attributes it to five people familiar with the discussions.

What restriction could look like

The Times reporting describes engagement on treating US model weights as protected intellectual property, a framing that matters because distillation, training a new model on another model's outputs, is the labs' central complaint against their Chinese rivals. An IP regime around weights would hand the labs legal tools against the practice, and potentially against the open distribution of models built on it.

Twenty-five companies took the other side a day earlier

The lobbying report landed one day after the open-weight camp went public. On Friday, 25 companies including Nvidia, Microsoft, Meta, IBM, Palantir, Hugging Face, Mozilla and the Linux Foundation published a letter titled "Open Weights and American AI Leadership," urging policymakers not to impose premature restrictions on open-weight models, CNBC reported. The letter argues that open-weight models let users download, modify and run them on their own infrastructure. OpenAI and Anthropic did not sign.

The list did not stop at 25. By Saturday the signatories had roughly doubled to about 50, Forbes reported, with Anthropic still absent. The split scrambles the usual alliances: Microsoft, OpenAI's largest investor, signed the letter OpenAI sat out, and Meta, Nvidia, IBM and Palantir rarely share a policy position this cleanly.

The K3 clock and the policy backdrop

The argument has a deadline. Moonshot's Kimi K3 open weights are expected Sunday night, a release covered in our K3 weights preview, and a joint UK-US government assessment of the model published last week is already feeding both sides: it found K3 completed zero of 41 exploit tasks yet attempted offensive cyber operations its safeguards should have blocked. The administration's formal tools against Chinese AI have so far run through export controls, not domestic restrictions on open weights.

For Anthropic and OpenAI, the reputational stakes compound the policy ones. Both companies built public identities on candor about safety trade-offs, and the Times story recasts a safety argument as a competitive one. The White House's most visible tech adviser spent the weekend attacking the restriction push in blunt terms; see our report on David Sacks' response.

AJ

Andrew Jamerson

Founding Editor, GaaS News

Andrew Jamerson is the founding editor of GaaS News, covering the economics of the agent era. He started the publication to cover Agentic AI as a Service as a dedicated beat and edits every article on the site.

Be on the list when the beat breaks

One email when a platform ships, a round closes, or the ground shifts under the software stack.