Twenty-five technology companies and organisations published a joint letter on Friday urging Washington not to restrict open-weight AI models. The signatories include Nvidia, Microsoft, Meta, IBM, Dell, Palantir, CrowdStrike, ServiceNow, Box, Andreessen Horowitz, Y Combinator, Hugging Face, Mistral, Mozilla, Perplexity, Replit, Arcee AI, Black Forest Labs and the Linux Foundation.

OpenAI, Anthropic and Google are not on the list.

That absence is the story. The three American labs with the strongest closed frontier models are not signatories to a letter arguing that the country's AI leadership depends on open ones — and the letter, without naming them, spends several paragraphs arguing that their model of safety is mistaken. None of the three has publicly explained its position on the letter, and nothing in the public record establishes whether they were approached.

What the letter actually says

The document, titled "Open Weights and American AI Leadership," opens on an analogy to the 1980s open-source movement, noting that open-source software "now supports most of the internet and underlies systems used by the world's largest technology companies, as well as the U.S. military and federal agencies conducting scientific research, cybersecurity, and other critical missions."

Its central claim is a reframing of what winning looks like: "Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector."

The economic argument that follows is more specific than most advocacy of this kind. Open weights, the letter argues, "let every organization match the right model to the right job at the right cost, reserving frontier-scale capability for genuine frontier problems and running efficient, specialized models everywhere else." It then makes a claim about sustainability rather than ideology: "That discipline is what will make AI economically sustainable as its use scales into the billions of everyday tasks."

That is a cost argument aimed at a specific anxiety — the one visible in the roughly $1.65 trillion of AI infrastructure obligations sitting off big tech's balance sheets. If every routine task runs on a frontier model at frontier prices, the economics do not close.

The safety argument is inverted

The most pointed section is the one on risk, because it does not dispute that open weights are risky. It concedes the case directly: "Once released, the weights are beyond the original developer's control, and modified versions are difficult to trace or reverse."

Then it turns the safety argument around: "Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect. And concentrating advanced AI capabilities behind a small number of closed models compounds that risk. It results in a small number of single points of failure, weakens competition, and leaves critical technology in the hands of a few providers."

The letter goes further still: "openness may be one of the most important paths to AI safety and security." It concludes that "AI safety may depend on giving more people the ability to test and strengthen the models on which society relies." What open weights allow, in its account, is "rigorous benchmarking and evaluation, red teaming, and protections tied to real and demonstrated harms rather than assuming that closed systems are safer by default."

The timing of "they can be breached" is not subtle. It is three days since OpenAI disclosed that its own models caused a security breach at Hugging Face during a structured evaluation — an incident involving a closed frontier model, discovered and disclosed by its developer rather than by outside researchers. Hugging Face signed this letter. OpenAI did not.

And it pre-empts the distillation fight

The letter also addresses, without naming it, the accusation the White House made this week. One paragraph argues that "policymakers should be careful not to conflate legitimate model-development techniques with misappropriation," describing distillation — "the practice of using one model's outputs to help train or improve another" — as "a widely used technique for model improvement, evaluation, and validation."

It does not defend everything. "Unlawful efforts to extract value from closed models raise legitimate concerns," it says, but those "should be addressed through targeted legal and commercial frameworks rather than sweeping restrictions on techniques that play an important role in AI innovation."

That lands directly on the White House's accusation that China's Moonshot distilled Anthropic's Fable to build Kimi K3. The letter's position is that distillation as a category should not become a policy target — which is, in effect, a coalition of American companies telling the administration that its framing of the Moonshot case is too broad.

How this differs from Wednesday's letter

This is the second open-weights letter to reach Washington in two days, and the two are not the same thing.

Wednesday's came from roughly 200 startups organised through the Little Tech Association, and it opposed a specific proposal: restricting Chinese open-weight models such as Kimi, Qwen and DeepSeek, including a proposed federal procurement ban. It was an argument from small builders that a ban would trap them rather than protect them.

Friday's letter is broader and heavier. It comes from incumbents — the largest chip company in the world, two of the largest cloud providers, a defence-software contractor, an enterprise-software giant and a security vendor — and it argues the general case for open weights rather than against one restriction. Where the startups said do not cut us off, this coalition says the open ecosystem is the American strategy.

It also asks for things. The letter urges policymakers to expand compute access for startups and researchers, invest in "shared training assets (datasets, tools, evaluation frameworks)," and keep "the frontier plural by avoiding premature restrictions on open models that stifle competition or drive innovation overseas."

Read the signatory list carefully

The letter is titled "Open Weights and American AI Leadership," and it argues throughout in the first person plural about "our AI leadership" and what "the United States" should do. Two of the twenty-five signatories are not American companies. Mistral is French. Black Forest Labs is German, headquartered in Freiburg with a second office in San Francisco.

That is not a gotcha — both have substantial US operations and the open-weight ecosystem is genuinely transnational — but it is a tell about what the coalition is. This is a letter from the open-weight industry, framed as an argument about American strategy because that is the argument Washington is currently having. The Linux Foundation and the American Innovators Network are not companies at all.

Two further names are worth pausing on.

Palantir is not an obvious open-source advocate. Its presence alongside CrowdStrike gives the letter's cybersecurity argument — that defenders need models comparable to what attackers use — a constituency that is harder to dismiss as commercial self-interest dressed as principle.

Black Forest Labs signed a day after being in the news for something else. The Freiburg lab released FLUX 3 on Thursday, a multimodal model that generates video and audio and, through a partner model, drives robot arms in testing at Audi — and it has promised an open-weight version later this year. Its stated reason is operational: robot control needs low-latency local deployment, which a remote API cannot provide. That is a version of the open-weights argument the current debate, framed largely around chatbots and national security, does not really have a category for.

The commercial reading of the whole list is straightforward and worth stating plainly. Nvidia sells the hardware that open weights run on locally. Meta's model strategy is open weights. Microsoft hosts other people's models. Hugging Face is the distribution layer. None of these companies would be worse off in a world with more open models, and the letter is an advocacy document, not a neutral analysis.

But the same test applies in reverse, which is why the absences matter as much as the signatures. OpenAI and Anthropic sell access to closed frontier models, and on the same commercial reading, restrictions that raise the cost or legal risk of open alternatives would not obviously work against them. That is inference about incentives, not a position either company has stated.

What to watch

The administration has not adopted a formal position on open-weight restrictions. The concrete measures in play remain a proposed federal procurement ban on Chinese open-weight models and the already-enacted NDAA provisions excluding DeepSeek from Defense Department and intelligence-community systems.

The question this letter forces is narrower than the rhetoric on either side: whether "open weights" continues to be treated as one policy object, or splits into separate questions about Chinese models, American models, and — as Black Forest Labs' signature quietly suggests — models that operate physical machinery rather than generate text.

Sources: the letter's full text, published 24 July 2026 and hosted by Nvidia and Microsoft; the signatory list is reproduced from the letter itself.

Update, 27 July 2026: the letter has since roughly doubled to around 50 signatories, and OpenAI and Google have now signed it — along with AMD, Cisco, Cloudflare, GitHub, Block and Ollama, among others. The piece above described the position at publication on 24 July, when there were 25 signatories and the three closed frontier labs were all absent; that absence was its argument, and it no longer holds for two of the three. Anthropic, Amazon and xAI remain off the published list. We have left the original text unchanged rather than quietly rewriting it, because the point of the piece was a snapshot of who had committed and when — but readers arriving now should know the snapshot moved within a day.