OpenAI and Anthropic push Washington to restrict Chinese open-weight AI models

OpenAI and Anthropic have declined to join a 20-plus-company coalition that on July 24, 2026 warned U.S. policymakers against broad restrictions on open-weight AI models, and have reportedly pushed for the opposite. Open-weight models are those whose trained parameters are released for anyone to download and run.

The coalition – including Nvidia, Microsoft, Meta, Dell Technologies, IBM, Palantir, Hugging Face, Mozilla, Mistral AI, Andreessen Horowitz and Y Combinator – argued in a joint letter that open models spur innovation, strengthen cybersecurity and keep the United States competitive. The two most prominent closed-model labs were absent from the signatories.

Both companies have reportedly urged the Trump administration to restrict open-weight models, citing alleged Chinese intellectual-property theft. The White House has accused Moonshot AI’s Kimi K3 model of distilling outputs from Anthropic’s systems. Anthropic CEO Dario Amodei has called open-source model scaling ‘a red herring’ and said it is ‘going down a very dangerous path.’

OpenAI CEO Sam Altman struck a softer public note, posting on X that he wants ‘the US to win in AI both in open source and proprietary models.’ Whether OpenAI ultimately signed the coalition letter after its initial absence drew attention is disputed and not independently confirmed.

The split has drawn a sharp charge from inside the administration. David Sacks, a Trump AI adviser, has accused leading closed-model labs of pushing restrictions that would amount to regulatory capture, entrenching incumbents against open-weight rivals.

The open question is whether the White House sides with the coalition or the two labs when it sets open-weight policy.