The deadline was set for last night. For the past one to two weeks, the Department of War has been demanding that Anthropic hand over its AI technology with no restrictions on how the Pentagon can use it in warfare. If Anthropic refuses, the threat is stark: its technology could be banned outright or declared a supply chain risk, with serious consequences for the company.
Anthropic has not complied.
What Anthropic won't surrender
Anthropic is the most outspoken AI lab on safeguards. The company has been explicit that it does not want its technology used for mass domestic surveillance or autonomous weapons. Those are precisely the applications the Pentagon's demand, as framed, would leave unrestricted.
The confrontation also carries baggage beyond the immediate dispute. There is pre-existing bad blood between Anthropic and the Trump administration, and observers close to the story sense that the Pentagon standoff is partly an expression of that accumulated tension rather than a clean disagreement over a specific contract.
Rivals closing ranks
What makes the moment genuinely unusual is who is standing with Anthropic. OpenAI, its direct rival, has drawn its own red line over how its technology would be used. Google is expected to face similar internal pressure, with employees pushing the company to take a comparable stand.
The two CEOs of Anthropic and OpenAI are not known for public solidarity. As Nick Wingfield, features editor at The Information, put it, they will not even hold hands at public gatherings where they are asked to. The unity here is striking precisely because of that rivalry.
The most plausible explanation is that it is driven by the engineers themselves. The people who built these systems went into AI wanting to cure diseases and solve hard problems. The prospect of their work being used for killing is something the workforce across these companies finds deeply uncomfortable. Every major AI lab needs to hire from the same pool of people, and that creates a shared pressure regardless of competitive dynamics.
A precedent, but not quite
The closest historical parallel is Project Maven. In 2018, Google chose not to renew a contract with the Pentagon after employees raised serious objections to the use of their technology in warfare settings. Google backed off.
Since then, the balance of power inside tech companies has shifted. Employers have become more resistant to petitions and protests, and employee activism has lost some of its leverage. But the underlying tension never went away, and Anthropic, with its leadership's explicit commitment to safeguards, is now carrying that argument forward at the institutional level rather than leaving it to internal dissent.
There is also a separate question that has not yet received much attention: whether AI is actually reliable enough to be used in the kinds of high-stakes military contexts the Pentagon is describing. That may yet become part of the debate in the days ahead.
How it ends
No one is certain. A compromise is possible. The Department of War wants Anthropic's technology, and that gives Anthropic some leverage. The two sides could find a way to communicate more clearly about what uses are and are not acceptable. Or the confrontation could escalate further, with consequences for Anthropic that remain hard to predict.
What is clear is that nothing quite like this has happened before. A government department issuing demands to a private AI company, with a hard deadline and a threat of effective exclusion, is new territory. How the government responds to an industry that is, for now, speaking with one voice will shape the relationship between Washington and Silicon Valley for years.