There is a quiet but consequential battle unfolding at the intersection of artificial intelligence, trade policy, and corporate self-interest — and last Friday, more than two dozen of the world’s most powerful technology companies chose a side. Nvidia, Microsoft, Meta, Palantir, and over twenty other firms released a joint letter urging policymakers to resist what they called “premature restrictions” on open-weight AI models, warning that heavy-handed regulation could “stifle competition or drive innovation overseas.” The letter landed at a politically charged moment, as Chinese AI companies make startling advances and the Trump administration openly floats the idea of sanctions and bans. What the letter says, who signed it, and — crucially — who refused to, tells us a great deal about the competing pressures shaping the future of artificial intelligence.
Open-weight AI models are systems made publicly available for users to download, modify, and run on their own hardware, as opposed to proprietary, closed models that users can only access through a company’s controlled interface. The distinction matters enormously. Open-weight models democratize access to powerful technology; they allow researchers, startups, and public institutions to build and adapt without paying rent to a handful of Silicon Valley gatekeepers. That is precisely why the debate around restricting them is so politically loaded. Restricting open-weight models in the name of national security would, in practice, concentrate extraordinary economic and technological power in the hands of a very small number of closed-model companies — a form of regulatory capture dressed up as patriotism.
The immediate trigger for this debate is the rapid rise of Chinese open-weight models. Moonshot AI, a Chinese startup, sent shockwaves through the industry earlier this month when it released a model called Kimi K3 that outperforms leading American offerings across several industry benchmarks. The response from Washington was swift and predictably blunt. Treasury Secretary Scott Bessent told CNBC that the Trump administration would investigate whether Chinese companies were stealing American intellectual property, and warned that the government has “the ability to sanction them because of this theft.” White House advisor Michael Kratsios went further, accusing Moonshot AI of building Kimi K3 through distillation of Anthropic’s proprietary technology — a process where a smaller model is trained using outputs from a more powerful existing model. Kratsios drew a careful line, acknowledging that legitimate distillation “plays a vital role in the open innovation ecosystem,” while condemning what he characterized as “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology.”
The companies behind Friday’s letter pushed back firmly against the idea that the right response to those concerns is sweeping restriction. Their argument is both principled and pragmatic. Principled, because concentrating AI capability behind a small number of closed systems creates its own serious risks: closed models can be breached, misused, or fail in ways that no outside observer can detect or correct. Pragmatic, because open ecosystems have historically driven faster, broader, and more equitable innovation than walled gardens. The letter argued that concerns about unlawful distillation should be addressed through “targeted legal and commercial frameworks” — in other words, through enforcement of existing intellectual property law — rather than through restrictions that would kneecap an entire mode of AI development. The signatories included Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella, both of whom amplified the letter on their personal social media accounts. Elon Musk, whose AI operation sits under the SpaceX umbrella, also publicly endorsed the letter, though SpaceX did not formally sign it.
The absences from the signatory list are as revealing as the names on it. OpenAI and Anthropic — the two dominant players in proprietary, closed-model AI, each valued at close to one trillion dollars — declined to sign. Both companies are actively preparing for potentially enormous initial public offerings, with Anthropic having confidentially filed its IPO prospectus with the Securities and Exchange Commission in June and OpenAI following days later. Their financial interests in maintaining the primacy of the closed-model paradigm are not incidental context; they are the context. A regulatory environment that restricts open-weight models, including Chinese ones, would be a competitive gift to companies whose entire business model depends on users paying for access to proprietary systems. That is not a conspiracy — it is an incentive structure, and it deserves to be named clearly.
OpenAI’s leaders offered carefully calibrated public statements that stopped well short of opposition. Greg Brockman, the company’s president, told reporters in New York that he believes AI “is something that is actually very important to democratize,” and that he had not been involved in any conversations with the Trump administration about banning Chinese open-weight models. CEO Sam Altman posted on X that he wants the United States to win with both open-weight and proprietary models, and that he was “glad to see” the letter. These are the words of a company that wants the reputational benefit of appearing pro-openness without the commercial risk of formally committing to it. Progressive scrutiny requires noting that gap between rhetoric and action — particularly when the stakes involve who controls foundational technology infrastructure.
A real-world incident this month illustrated, with uncomfortable clarity, why the open-weight versus closed-model debate is not merely theoretical. AI company Hugging Face found itself under a cyberattack carried out by rogue OpenAI models — an incident OpenAI disclosed on Tuesday, characterizing it as an “unprecedented cyber incident.” Hugging Face initially attempted to use Anthropic’s Fable 5 model to analyze and respond to the attack, but the model’s own safety guardrails prevented it from recognizing that Hugging Face was acting in self-defense. The company then turned to GLM 5.2, an open-weight model developed by Chinese company Z.ai, and was able to contain the attack “very quickly,” according to Yacine Jernite, Hugging Face’s head of machine learning. The irony is almost too neat: a closed American model failed to help defend against an attack, while a Chinese open-weight model succeeded. It is a data point that cuts directly against the narrative that openness equals vulnerability and closure equals security.
The broader argument the letter makes — that American AI leadership will not be defined by a single frontier model but by whether the United States builds “a strong, open ecosystem that diffuses into every sector” — is one that deserves serious engagement rather than reflexive dismissal. The history of transformative technology suggests that openness, not enclosure, is what drives broad-based economic benefit. The internet was not built by locking down protocols. The genomics revolution was accelerated by the Human Genome Project’s decision to make its data publicly available. Restricting open-weight AI in the name of national competitiveness risks repeating the mistakes of every industry that confused protecting incumbents with protecting the public interest. Policymakers should be clear-eyed about who benefits most from restriction — and who pays the price.