The Open-Weights Letter Doubled to 50 Signers in a Day. OpenAI Signed After the Backlash. Anthropic Still Won't.

TL;DR

On Friday, 25 American companies published Open Weights and American AI Leadership, an open letter asking Washington to keep "the frontier plural by avoiding premature restrictions on open models that stifle competition or drive innovation overseas." Within about a day the signature block doubled to 50 names, including Google and, only after its absence went viral, OpenAI. Anthropic and Amazon are still not on it. The sting in the tail: Axios and the New York Times report that the two biggest closed labs spent the same stretch quietly pressing Washington in the opposite direction.

What the letter actually says

The letter, dated July 24 and hosted as a PDF on Nvidia's own site, defines open-weight models as AI models "anyone can download, inspect, modify, and run on their own infrastructure" and argues they are how the US wins the AI era: cheaper access for startups and universities, rivalry across chips, clouds, and applications instead of a captive API market, and no lock-in to a single provider. It asks policymakers for compute access for researchers, shared datasets and eval frameworks, and restraint on restrictions.

The safety section is the part worth reading twice. "Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect," the letter argues, adding that concentrating capability behind a few closed systems creates "a small number of single points of failure." That is a direct inversion of the closed-lab framing: openness recast as the safety strategy, and closure as the risk.

Then there is the distillation paragraph, which reads like it was drafted with a specific grudge in mind. Distillation, using one model's outputs to help train or improve another, works a bit like learning a rival chef's menu by ordering takeout every night: you never see the kitchen, but enough meals get you most of the recipes. The White House accused Moonshot AI of doing exactly that to Anthropic's Fable 5 earlier this month. The letter calls distillation "a widely used technique for model improvement, evaluation, and validation" and says abuse should be handled "through targeted legal and commercial frameworks rather than sweeping restrictions on techniques" that drive AI innovation.

Nine days, one reversal

The sequence is the story. David Sacks, who ran AI policy for the Trump administration, spent mid-July arguing that Chinese open-weight releases like Kimi K3 were pulling developers onto Chinese stacks, then turned his fire on the closed labs, saying the "leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition," a move he described as regulatory capture, per Gizmodo.

On July 22, Axios reported OpenAI and Anthropic were aligned in pressing Washington on open-model risks. Two days later the letter dropped without either of them, the omission got noticed loudly on X, and OpenAI added its signature the same day. CEO Sam Altman: "I want the US to win in AI both in open source and proprietary models, and I am glad to see this." Then on Sunday the New York Times put the split on its front page: the two closed-model leaders publicly nodding at openness while lobbying regulators to restrict open-source rivals.

Anthropic, to its credit, is not pretending. It never signed, and CEO Dario Amodei has long called open source "a red herring" in AI. You can disagree with the position; at least it matches the lobbying.

Who signed, and who didn't

Nvidia CEO Jensen Huang promoted the letter with his first-ever post on X. The man shipped CUDA before he shipped a tweet. Alongside Nvidia, Microsoft, Meta, and IBM, the roster includes essentially every layer of the self-hosted AI stack: Hugging Face, Ollama, Mistral, Black Forest Labs, Nous Research, Prime Intellect, plus GitHub, Cloudflare, Mozilla, the Linux Foundation, Andreessen Horowitz, and Y Combinator. Fifty names as of Saturday, per the PDF's own signature block.

The conspicuous holdouts are Anthropic and its biggest investor, Amazon, a pairing Forbes flagged as the story inside the story. Google joined the second wave, which costs it little: Gemma is already one of the most-downloaded open families on Hugging Face. Anthropic is the only frontier lab on either side of the fight with no open-weight release at any size.

Why builders should care

Per Gizmodo's summary of the fight, the proposals circulating in Washington include putting Chinese AI labs on the Commerce Department's Entity List, executive orders holding US companies accountable for hosting Chinese models, and official warnings about backdoors in Chinese weights. None of that is enacted, but "accountable for hosting Chinese models" is a phrase that lands directly on Hugging Face mirrors, on inference providers, and eventually on the GPU box in your rack.

The timing explains the heat. GLM 5.2 sits near the coding frontier at open-weight prices, DeepSeek ships variants people actually run on home hardware, and Moonshot's Kimi K3 weights are due for public release on July 27, tomorrow. The open frontier the letter defends is, increasingly, a Chinese one, which is exactly the awkward fact both sides of this letter are arguing over.

Caveats

• The letter is advocacy, not analysis. Most signers sell picks and shovels for open weights: Nvidia sells the GPUs, Hugging Face hosts the downloads, Ollama wraps the runtimes.
• The signature list is a moving target. It went from 25 to 50 in roughly a day; the PDF on Nvidia's site is the canonical roster and may grow again.
• The lobbying claims come from journalism, not public filings. Axios and the New York Times describe private pressure; neither OpenAI nor Anthropic has published its asks.
• Nothing here is law. No bill text or executive order has been released; the Entity List and hosting proposals are reported as under discussion.

Key Takeaways

• A letter titled Open Weights and American AI Leadership, published July 24 and hosted by Nvidia, urges Washington to avoid "premature restrictions" on open-weight models.
• Signatures doubled from 25 to 50 in about a day, and the list now covers nearly the entire self-hosted AI stack, from Hugging Face to Ollama to Mistral.
• OpenAI signed only after its absence went viral; Axios and the New York Times report both OpenAI and Anthropic privately lobbied for restrictions on open-source rivals.
• Anthropic and Amazon remain the notable holdouts, and Anthropic is the only frontier lab with no open-weight release at all.
• Proposals under discussion include Entity List designations for Chinese labs and accountability for hosting Chinese models, which would directly touch mirrors, inference providers, and self-hosters.
• Nothing is enacted yet. The fight is over whether the default stays "downloadable" or becomes "licensed."

Sources: Open Weights and American AI Leadership (letter PDF), Forbes, Gizmodo, Axios, AI Weekly