2026-02-27 08:50
publication
2026-02-27 08:50
Anthropic, the creator of the leading AI model Claude, on Thursday rejected the Pentagon's ultimatum, which demanded that the company abandon restrictions on the use of its models for mass surveillance of Americans and autonomous weapons. The Pentagon is threatening the company with sanctions or legal coercion.
The company's decision was announced by its CEO, Dario Amodei, a day before the ultimatum given by Pentagon chief Pete Hegseth. As he wrote in a statement, although his company supports cooperation for the country's national security and has not yet made any ad hoc decisions to limit the use of its products, it cannot abandon the demands that the contract with the Pentagon include protections against the use of AI for mass internal surveillance and autonomous weapons that make decisions without human intervention (called “killer robots” by critics).
Amodei said he was not completely against autonomous weapons in the latter case, but noted that such use would be dangerous given the current state of technology.
“We have offered to work directly with the Ministry of War on research and development to improve the reliability of these systems, but the Ministry has not accepted this offer. Furthermore, without proper oversight, fully autonomous weapons cannot be relied upon to exercise the critical judgment that our highly trained, professional soldiers demonstrate on a daily basis. They must be deployed with appropriate safeguards, which are currently lacking,” the Anthropic chief wrote.
Amodei confirmed earlier media reports that the Pentagon had threatened his company not only with the removal of the company's products from Pentagon systems, but also with formal recognition as a “supply chain threat”, which is a sanction previously reserved for companies from countries hostile to the US. This designation requires all U.S. government contractors to terminate cooperation with the company – which includes almost all of the largest American companies. The Anthropic founder added that the Pentagon also threatened to use the Defense Production Act to force the removal of the protections.
“These last two threats are inherently contradictory: one defines us as a security threat, the other defines Claude as essential to national security,” Amodei noted. “Nonetheless, these threats do not change our position: we cannot in good conscience agree to their demand.”
Anthropic is the only company whose products are used in classified government resources, and according to the Washington Post, its Claude model was also used during the operation to capture Venezuelan leader Nicolas Maduro. The value of the current contract with the Pentagon is $200 million. Anthropic is currently valued at $380 billion.
The company's announcement comes a day after Anthropic softened its previous stance regarding the security of its AI models: while so far, the company has firmly said that it will not stop work on a model if its abilities exceed the company's ability to control and ensure its security. Now, the company says it can't afford it if its competitors don't use the same safeguards. A cautious approach to security has been the main feature emphasized by the company so far.
Pentagon spokesman Sean Parnell, on the one hand, denied on Thursday that the Pentagon intended to use AI for mass surveillance – which, as he stressed, is not allowed by law – or autonomous weapons, but on the other hand, he demanded that the company abandon its demands.
“Here's what we're asking: Allow the Pentagon to use the Anthropic model for all lawful purposes,” Parnell wrote on Platform X. “We will not allow any company to dictate how we make operational decisions. They have until 5:01 p.m. EST on Friday to make a decision. Otherwise, we will terminate our relationship with Anthropic and consider it a supply chain threat to DOW.” – he added.
From Washington Oskar Górzyński (PAP)
osk/ sp/
