Dario Amodei, CEO of Anthropic, said Monday that the company has never tried to prohibit open-weight AI models, pushing back on regulators weighing new rules and on rivals who say Claude has used safety and security arguments to defend its position.

The clarification lands as Washington considers how far new restrictions should go. The Senate has introduced a bill that could lead to a federal limit on open-weight models, a category of systems developers can download and run on their own machines. The outcome could shape which labs gain a competitive edge, what kinds of models companies can bring to market, and how dominant closed-model providers become, which is why the dispute has become part policy fight and part industry positioning.

The chokepoints Amodei wants regulated instead

In his blog post, Amodei did not call for a blanket ban. Instead, he outlined three narrower policy priorities for lawmakers.

First, he wants tighter controls on advanced chip technology and computing equipment shipped to authoritarian countries such as China. Second, he wants strict enforcement against industrial “distillation,” the practice of training a new model using the outputs of a more advanced one. Third, he wants mandatory testing of powerful models to help ensure their safety.

Amodei said a ban would miss the real threat because, in his words, “bad actors are unlikely to be legitimate US businesses.” In comments to Axios, he acknowledged that banning Chinese open-source models could help reduce competition for labs like his own, but said that “that has never been my goal.” Bloomberg Law reported that his main concern is authoritarian states developing AI that outperforms US products.

The letter Anthropic would not sign

Amodei’s remarks followed a public refusal to sign an industry letter. More than two dozen companies, including Nvidia, Meta, and Microsoft, had signed an earlier letter titled “Open Weights and American AI Leadership,” urging lawmakers not to impose restrictions on the technology, according to Business Insider. Over the weekend, OpenAI, Google, and SpaceX joined the effort, leaving Anthropic as the only major firm not to sign by Monday.

The split quickly drew attention in Silicon Valley. David Sacks, the venture capitalist and former Trump administration AI adviser, warned on X that the industry should “watch these guys like a hawk.” He wrote:

The entire tech industry (save for Anthropic) has come out in favor of open source AI. So what happens next? Will Anthropic change its lobbying efforts? Not likely. Now the gaslighting begins: “Nobody is trying to ban open source.” “We just want to limit who can use it.” “We… — David Sacks (@DavidSacks) July 26, 2026

Benchmark’s Bill Gurley responded that Anthropic was protecting its “corporate economic strategy.” Kai-Fu Lee, founder of Chinese open-model startup 01.AI, said the company that did not sign was “far more interesting than who did.”

From shared concern over China to a public split

The disagreement is more visible now, but the underlying issue is not new. Closed-model competitors have been aligning in Washington for some time. In July, Axios reported that OpenAI and Anthropic joined forces to brief lawmakers on powerful Chinese open-weight models, a move that drew criticism from Sacks and others who described it as a potential example of regulatory capture that could entrench large companies.

The latest debate was sparked by Kimi K3, an open-weight model from Chinese startup Moonshot that rivaled US models at a much lower price. Some accused Moonshot of using Anthropic’s Fable model to train Kimi K3 through distillation. Treasury Secretary Scott Bessent then proposed sanctions against Chinese models found to be copying US systems. After that, OpenAI and Google, two of Anthropic’s biggest competitors, signed the open letter, leaving Anthropic isolated and prompting questions about whether the company favored restrictions to protect its business interests.

What Amodei accepted and what he rejected

Amodei tried to strike a careful balance. He said open models improve accessibility, encourage competition, and give customers more control. He also described open models without dangerous capabilities as “a public good.”

At the same time, he rejected the letter’s central safety claim. He said he does not believe open weights make it easy for defenders to build protections, or that broader access benefits defenders more than attackers. In his view, any capable model could be used to weaponize pandemic-level viruses, while building defenses would take far longer.

The industry divide is now highly visible. Nvidia used the moment to announce the Open Secure AI Alliance, backed by Microsoft, SpaceX, Hugging Face, and others. The group said open technologies are protected by law and argued that a total ban would hand too much power to a small number of closed-model providers.

Legislators have yet to establish the standards that will determine the outcome of the debate.