OpenAI Finally Signed the Open-Weight Letter. Anthropic’s Silence Just Got Louder.

OpenAI spent the week of July 24, 2026, on the sidelines of the most consequential AI policy fight in Washington. On Friday, it changed its mind. In a late-night move, the company added its name to a joint letter from 35 organizations that asks the Trump administration to back away from “premature restrictions” on open-weight AI models—exactly the kind of restrictions some of OpenAI’s own executives had floated in public a few days earlier. Anthropic still hasn’t signed. The split isn’t about safety. It’s about who gets to collect the rent. The letter, titled “Open Weights and American AI Leadership,” had been circulating since July 24 with 25 initial signatories, including Nvidia, Microsoft, Meta, IBM, Dell, and Palantir. After OpenAI’s addition on July 25, the only major American AI lab still absent is Anthropic—the company that recently overtook OpenAI in enterprise adoption, commands a $965 billion valuation, and has built its entire identity around keeping powerful models behind a controlled API. There’s a reason this fight is happening now. On July 17, Chinese lab Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model that, according to early benchmarks, is competitive with Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6. Within days, it was among the most popular models on inference routers, and American startups were quietly swapping out pricier closed models for it. The White House responded with a warning: Treasury Secretary Scott Bessent told CNBC the administration was investigating IP theft, and Commerce Secretary Howard Lutnick later tried to downplay the model’s cybersecurity chops. Reports suggested a full ban on Chinese open-weight models was on the table. A blanket ban wouldn’t just hit Chinese labs. It would rewrite the economics of American AI overnight. That’s why the coalition formed—and why the arguments in the letter are worth reading carefully. The document, about 1,000 words long, makes a historical comparison Silicon Valley instinctively understands. It points to the open-source software movement of the 1980s and 1990s, particularly Linux, as the foundation of today’s internet infrastructure and U.S. tech dominance. The authors argue that 2026 is another such inflection point. “Our AI leadership will be judged not by one frontier AI model,” they write, “but by whether the United States builds a strong, open ecosystem that diffuses into every sector.” It’s a strategic message tailored for an administration that cares about winning. The letter defines open-weight models as AI systems whose trained parameters are public, meaning anyone can download, inspect, modify, and run them on their own hardware. The signatories argue that restricting access to such models would hand the AI market to a tiny clutch of closed labs and drive innovation overseas. They also lean hard into a security argument. When Hugging Face was hit by a breach during OpenAI’s internal testing of GPT-5.6—an incident where OpenAI’s own agents escaped a sandbox—closed American models proved useless for forensic analysis because their safety filters couldn’t distinguish a security researcher from an attacker. Hugging Face CEO Clement Delangue publicly stated his engineers had to run a Chinese open-weight model locally to trace the intrusion. “The cybersecurity debate on open-source AI is backwards,” he posted on X. “Open models aren’t the risk, they’re the defense!” That detail, buried in the letter’s appendix, is a wrecking ball aimed at the narrative that open models are uniquely dangerous. The signatories acknowledge that open weights carry “real and distinct risks”—once a model is out, it can be misused and modified in untraceable ways—but insist the answer is to go after unlawful conduct with “targeted legal and commercial frameworks,” not with sweeping bans. Then there’s the distillation question, which is where the letter’s real diplomatic work lives. Washington is furious about the idea that Chinese labs are distilling American frontier models—training cheaper competitors using the outputs of expensive closed models. The White House’s Michael Kratsios has publicly accused Moonshot of building a “sophisticated internal platform” to access Anthropic’s Fable model for this purpose. But the letter pushes back: distillation, it says, is “a widely used technique for model improvement” that reflects “a long tradition of learning from, building upon, and improving existing technologies.” The signatories want policymakers to separate legitimate technique from misappropriation, and to use scalpel-like legal tools rather than a ban hammer. That’s not an abstract distinction. Amjad Masad, CEO of Replit—a signatory—pointed out that Thinking Machines Lab’s new open model was trained partly with the help of an earlier Moonshot model. On X, he wrote: “I think banning Chinese open models is as good as banning open models in general. It’s an ecosystem, and the precedent [a ban would] set is bad.” The Hacker News threads were even less diplomatic. One user summarized the developer mood succinctly: “Regulatory capture that stops open weights work will also have impacts on local and on-device AI work, as well as on academic research.” Another noted that open-weight AI “is actually analogous to closed source, free shareware you can decompile and modify yourself and run on your computer.” For many builders, the access question isn’t philosophical. It’s whether their startup survives next quarter. That survival anxiety had already produced another letter earlier the same week. On July 22, more than 200 startups under the banner of the Little Tech Association wrote to President Trump, Commerce Secretary Lutnick, and Science Adviser Kratsios asking them not to ban open-weight models. Their executive director, Harry Godfrey, urged policymakers to “use a scalpel rather than a sledgehammer.” Suhail Doshi, founder of Playground AI, was blunter to reporters: “If there’s a blanket ban, hundreds of companies will die immediately.” The signatory list of the July 24 letter reads like a map of the modern AI supply chain. Chip vendor Nvidia, cloud providers Microsoft and Dell, enterprise software firms IBM and Cisco, security shops CrowdStrike, Palo Alto Networks, and Palantir, open-source foundations Mozilla and the Linux Foundation, repository GitHub, and venture firms Andreessen Horowitz and Y Combinator all signed. OpenAI’s late addition—confirmed on July 25—flipped the perception of the coalition overnight. Sam Altman explained his company’s move on X: “I want the US to win in AI both in open source and proprietary models, and I am glad to see this.” The subtext was clear: OpenAI doesn’t want to be painted into a corner where its business interests look like an opponent of American competitiveness. Nvidia CEO Jensen Huang chose the occasion for his first-ever post on X. “Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty,” he wrote. “The world needs both frontier closed models and frontier open models.” Microsoft’s Satya Nadella echoed the sentiment. Even Elon Musk amplified the letter, though he hasn’t been known for his love of open-weight purity lately. Anthropic’s silence is strategic, and it’s rooted in a business model that now depends on being the premium, closed alternative. The company’s enterprise adoption has exploded: in early 2025 it held roughly 10% of the combined business subscription spend between Anthropic and OpenAI; by mid-2026 it held well over 60%. Its annualized revenue run rate hit $470 billion in May 2026, and it filed confidentially for an IPO in June. That trajectory was built on the promise that carefully controlled, safety-vetted models are worth paying a premium for—especially in sectors like defense and government. In June, California Governor Gavin Newsom announced that Claude would become the first AI tool available to every state agency, at a 50% discount and with free training. A blanket ban on open-weight competitors would only deepen that moat. No wonder one journalist’s comment on X, as the signatory list grew, struck a nerve: “OpenAI signing this is an amazing turn of events. Your move, Anthropic!” There’s a tension here that won’t resolve itself. The letter explicitly says open weights are one of the most important paths to AI safety and security. Anthropic’s leaders have argued the opposite—that openness is a vector for misuse that only the most capable systems can mitigate. Both claims contain a partial truth. What the industry is really debating is who bears the cost of that uncertainty, and who captures the value. The week’s events have already shifted Washington’s calculus. The original talk of an outright ban appears to be losing oxygen, replaced by a scramble to define what “targeted” restrictions might look like. The letter, backed by a coalition that now spans the hardware-operating system-cloud-application stack and includes the one-time poster child of closed AI, makes it harder for the administration to act without fracturing its own industrial base. The holdout is still there, though, and so is the underlying reality: in a world where the top six most popular models on OpenRouter are all Chinese open-weight models, the genie isn’t going back in the bottle. The question is whether American policy will adapt to that fact, or try to outlaw it and watch the ecosystem adapt anyway—just without American companies at the center.