Anthropic Spent 48 Hours Correcting a Record Nobody Will Forget

The letter already had 50 signatures by the time Dario Amodei sat down to explain why Anthropic did not sign it.

Jensen Huang's open-weights letter doubled its signatories to 50 in a single day, including OpenAI and Google. Google and OpenAI signed on over the weekend after originally abstaining. Anthropic still has not. By Monday morning the narrative was fixed: Anthropic stood alone.

So Amodei published. "To summarize my and Anthropic's position, we have not and are not advocating for a ban on open-weights models as a category," he wrote. He said Anthropic would instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed. He drew a clear distinction between smaller open models, which provide real benefits, and the most advanced frontier systems, where capabilities might enable serious harm and where, once released, there is no undo button.

Every word of that is defensible. None of it matters for the next six months of DC conversations.

The First Voice Sets the Frame

The trigger was Kimi K3. A Chinese open-weight model rattled Silicon Valley by approaching U.S. frontier performance at a fraction of the cost. A Chinese lab shipping a near-frontier model as open weights forced every closed-model company to pick a side publicly.

OpenAI and Google picked first. That decision was not a principled stand; it was a first-mover advantage move. Supporting a letter verbally gives you the PR benefit of alignment without the binding commitment of your logo on a public document. They hedged, then committed. Anthropic neither hedged nor committed.

Translation of Amodei's statement: we agree frontier open-weights is dangerous, we disagree with the letter's safety claims, and we are not joining the coalition that said it first.

The problem is the sequence, not the substance.

What Isolation Costs

Anthropic's absence fueled accusations, including from White House AI adviser David Sacks, that the company was using safety concerns to defend its own business model. When your harshest critic is the sitting White House AI adviser, you do not fix that with a blog post.

Anthropic has a commercially specific grievance here that makes the safety position harder to read as purely principled. Anthropic sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs alleging that China's Alibaba had carried out the "largest known distillation attack" against it. Anthropic accused Moonshot of using 3.4 million Claude exchanges to train its models through distillation, and Kimi K3 now benchmarks within a few points of the models named in that complaint. The company with the most direct commercial injury from the exact models under debate is also the company most resistant to the letter defending those models.

That is a difficult optic to explain in a Senate hearing room.

Amodei's distinction between opposing a ban and endorsing the letter's specific claims is technically real. He agrees that open models can expand access, competition and customer control, but rejected the letter's claim that open-weight models necessarily make AI safer or give cyber defenders an advantage over attackers. That is a legitimate disagreement with one paragraph of a multi-page document.

In Washington it reads as: Anthropic would not sign.

By July 27, Anthropic was the only leading frontier AI company that had not endorsed the industry-wide campaign defending open-weight AI models from regulatory restrictions. That sentence is the one that will be quoted in staff briefings, not Amodei's nuanced explanation of frontier-scale risk thresholds.

If Congress writes a definition of "frontier" into statute, that definition will carry the fingerprints of the companies that shaped the debate loudest and earliest. Anthropic opposed the same risks, arrived with a better argument, and still lost the room.

Being right too late is the oldest mistake in Washington.

What to Watch

Watch whether Congressional staff cite the letter by name in hearing questions or bill language over the next 30 days. If yes, OpenAI and Google have already cashed the first-mover dividend.

Watch for Anthropic announcements around smaller open-weights releases as a counter-move to rebuild credibility with the open-source community.

Watch whether Amodei's safety-testing proposal—mandatory evaluation for all sufficiently capable models, open and closed—picks up any co-signers from the letter coalition. If it does, Anthropic gets back in the game. If it does not, the company will have spent July correcting a record while its rivals wrote the next one.