25 Tech Companies Push White House to Protect Open-Weight AI as Companion Tools Hang in the Balance
More than 20 major technology companies, including Nvidia, Microsoft, Meta, and Hugging Face, signed an open letter on Friday urging the Trump administration not to impose broad restrictions on open-weight AI models, warning that such limits would undermine competition and push users toward Chinese AI systems. The letter, sent to Office of Science and Technology Policy Director Michael Kratsios and shared first with POLITICO, represents the industry's most coordinated defense of open-weight AI against a rising regulatory threat.
The development comes as the White House weighs restrictions on advanced Chinese open-weight models over national security concerns, with Axios reporting that the administration is considering banning models like K3. However, according to a separate Politico report, the Department of Commerce has not moved toward that step anytime soon. The tension has placed the future of open-weight AI at the center of US technology policy.
What the Open Letter Says
The July 22 letter — co-signed by 25 companies including Nvidia, Microsoft, Meta, IBM, Dell Technologies, Palantir, Hugging Face, and Mistral — argues that American developers should remain free to publish model weights unless the government can demonstrate that publication itself creates a distinct material risk. The signatories call for safeguards tied to capability, users, and deployment rather than economy-wide restrictions.
"Federal policy should preserve the ability of American developers to release open-weight models while addressing specific security risks through safeguards tied to capability, users, and deployment," the letter states, according to a copy obtained by POLITICO. Nvidia CEO Jensen Huang, who joined X last month, used his first post to promote the three-page policy letter on Friday.
The Companion AI Stake
For AI companion users and NSFW AI developers, this debate carries immediate practical stakes. Open-weight models — from Meta's Llama series to Mistral, Qwen, and Nemotron — form the technical backbone of locally-run, self-hosted companion AI. Unlike API-gated models from OpenAI or Anthropic, which impose strict NSFW content policies, open-weight models can be run without filtering, making them the primary option for uncensored companion experiences, roleplay, and adult content generation.
If broad restrictions were placed on open-weight models — particularly Chinese models like Qwen from Alibaba, which are widely used in the companion AI community — developers would lose key options for cost-effective, high-quality local inference. The debate also intersects with existing state-level AI companion laws like California's SB 243, which already requires age verification and safety protocols for companion chatbot operators regardless of underlying model source.
How We Got Here
The NTIA's July 2024 report on widely available model weights found that evidence was not sufficient to conclude that blanket restrictions are warranted, but recommended monitoring, audits, and risk thresholds. The June 2, 2026 executive order on AI took a voluntary approach, asking frontier AI companies to give the government up to 30 days of pre-release access to models judged to have advanced cyber capabilities, while explicitly rejecting mandatory licensing.
Now, with advanced open-weight models from China — particularly DeepSeek's R1 and the Qwen series — matching or surpassing US closed models on key benchmarks at lower cost, the pressure for restrictions has intensified. The New York Times reported on July 21 that certain Trump administration members advocate for restrictions on foreign models.
What Comes Next
The most plausible near-term US approach, based on the evidence gathered across multiple reports, is targeted controls on high-risk deployments and foreign-adversary-controlled systems rather than a general prohibition on publishing weights. But the companion AI industry should watch this space closely: any broad restriction on open-weight models would reshape the availability of models that developers — and users — depend on for locally-run, unfiltered AI experiences.