The Claude AI maker’s CEO, Dario Amodei, says the company will not “in good conscience” grant the Pentagon unrestricted control over its AI.

In a statement released on Thursday, Amodei said that while it supports US defense, it rejects the use of its technology for mass surveillance and fully autonomous weapons, citing democratic and safety concerns.

It follows the AI firm's high-stakes standoff with the Pentagon, which has reportedly given Anthropic until 5:01 p.m. on Friday to remove restrictions on the use of Claude, including guardrails prohibiting mass surveillance of Americans and autonomous weapons, or risk being designated a “supply chain risk."

Amodei said the company “cannot in good conscience accede” to the Department of War’s requirement that AI contractors permit “any lawful use” of their technology, including applications Anthropic has explicitly barred.

The AI head stressed that the company supports US national security efforts for mission-critical applications and acknowledged that it is the Department of War, not private companies, that makes military decisions.

The company also noted it had taken steps it described as costly to its own business interests to protect US advantages in AI, including cutting off access to firms linked to Beijing and advocating for stronger chip export controls.

But Anthropic drew firm boundaries around two uses it says undermine democratic values or exceed current technological reliability: “Mass domestic surveillance” and “fully autonomous weapons.”

“Using these systems for mass domestic surveillance is incompatible with democratic values,” the company said, warning that AI could compile “a comprehensive picture of any person’s life – automatically at a massive scale.”

Anthropic’s position puts it at odds with other firms that supply the US government agencies with tools that can be used in this way, including data analytics firm Palantir, which won a $30m contract to develop "ImmigrationOS," an AI-backed platform designed to identify, track, and deport so-called “noncitizens.”

Curious what others think about this story? Contribute your thoughts to the debate below.

Anthropic’s other pushback was on the use of autonomous weapons – weapons that select targets without human oversight. Currently, AI is not only untrustworthy but also very dangerous when unsupervised. In military operations, it can also be used to dehumanize operations by offering gamified experiences for officers and soldiers, and shifting personal responsibility.

Acknowledging this risk, the AI firm said that while partially autonomous systems play a role in modern warfare, “today, frontier AI systems are simply not reliable enough to power fully autonomous weapons."

“We will not knowingly provide a product that puts America’s warfighters and civilians at risk.”

Anthropic rivals, OpenAI and Google, have reached agreements to let the Pentagon use their AI tools under the “any lawful use” framework, including in classified settings, while xAI recently struck a deal to allow its AI (Grok) to be used by the Department of Defence, effectively aligning to the Pentagon’s terms.

“Supply Chain Risk” pushback

The statement confirms earlier reports that the Department of War threatened to remove Anthropic’s systems from government networks and potentially designate the company “a supply chain risk” if it maintains those safeguards.

Officials have also suggested invoking the Defense Production Act to compel changes, Anthropic said.

The company argued that those threats are contradictory – labelling it both a potential security risk and an essential supplier – and reiterated that its position will not change.

Sean Parnell, the Pentagon’s top spokesman, said earlier on social media that the military “has no interest in using AI to conduct mass surveillance of Americans (which is illegal) nor do we want to use AI to develop autonomous weapons that operate without human involvement.”

Anthropic’s stance comes amid reports on Tuesday that it is scaling back its voluntary AI safety pledge, a move that may complicate its argument for maintaining stricter safeguards in defense applications.