Defense Secretary Pete Hegseth has demanded unrestricted access to Claude for all military applications, putting Anthropic's safety-first positioning on a direct collision course with the US government
The confrontation between Anthropic and the Pentagon has reached a crisis point. Defense Secretary Pete Hegseth summoned Anthropic chief executive Dario Amodei to a meeting at the Pentagon on Tuesday and delivered an ultimatum: remove the restrictions the company places on how its AI can be used by the military, or be cut from theDoD's supplier network entirely.
The deadline is Friday evening
According to a source who described the encounter, the atmosphere was civil. Amodei reportedly thanked Hegseth for his service. But beneath the pleasantries, both sides laid out positions that are difficult to reconcile.
Anthropic maintains that certain applications fall outside what its models can do reliably and responsibly. The Pentagon's stated position is that it should face no restrictions whatsoever on how it uses the technology it has contracted to deploy.
What Anthropic will and will not do
The company has identified two firm limits. The first concerns autonomous weapons: Anthropic will not allow its tools to make final targeting decisions in military operations without human intervention in the loop. The second concerns domestic surveillance at scale. Both represent what the company privately describes as red lines.
A Pentagon official who briefed the BBC was careful to say that the current dispute is not specifically about autonomous weapons or mass surveillance.
The precise nature of the disagreement has not been made public, which leaves the core question unanswered: what exactly does the Defense Department want to do with Claude that Anthropic is currently preventing?
The leverage the Pentagon holds
The threat Hegseth brought to the meeting is not merely contractual. A senior Defense official has intimated that if Anthropic does not comply by Friday, Hegseth intends to invoke the Defense Production Act, a wartime authority that could compel the company to grant unrestricted access on national security grounds.
Simultaneously, the Pentagon would designate Anthropic a supply chain risk, a label with serious consequences for a company whose business model depends on institutional trust.
Anthropic was among four AI companies awarded Pentagon contracts worth up to $200 million each last summer, alongside Google, OpenAI and Elon Musk's xAI. Defense official Emil Michael has previously stated the department's expectation that all four allow the Pentagon to deploy any of their models across all lawful use cases without restriction.
How the relationship deteriorated
Observers say the immediate trigger for the breakdown was the disclosure that Claude was used during the operation that led to the capture of former Venezuelan president Nicolás Maduro in January, apparently through Anthropic's partnership with Palantir, which holds its own Pentagon contracts.
Anthropic had not sanctioned that specific use, and the episode raised questions about whether the company's restrictions were enforceable in practice once its technology entered classified government networks through third parties.
Anthropic had been the first technology company cleared to operate inside the Pentagon's classified systems. That distinction, once a mark of the company's institutional standing, now sits awkwardly against the current confrontation.
Analysts framed the problem plainly
The two sides need to reach a resolution, she said, because the people the military asks to serve in the field deserve every available advantage.
The question is whether that resolution requires Anthropic to abandon the safety boundaries that define its public identity, or whether the Pentagon is willing to accept something short of unconditional access.
Friday will provide the answer.