Anthropic has refused a Pentagon request to remove safety restrictions from its AI systems, even after being warned it could be labeled a “supply chain risk” and lose a $200 million Defense Department contract. The dispute centers on Anthropic’s decision to keep safeguards that prevent its technology from being used for fully autonomous weapons or mass domestic surveillance.

Pentagon spokesperson Sean Parnell said the department does not intend to use AI for mass surveillance of Americans or for weapons that operate without human control. However, he stated the Pentagon wants the ability to use Anthropic’s model for all lawful purposes and gave the company a deadline to comply or face termination of the partnership.

Anthropic CEO Dario Amodei reaffirmed the company’s opposition to removing the safeguards, arguing that advanced AI systems are not reliable enough for life-or-death military targeting. A source close to the company clarified that Anthropic was not accusing the Pentagon of planning such uses, but was making a product safety judgment. The source warned that unpredictable AI behavior could lead to friendly fire, mission failure, or unintended escalation. They also raised concerns that current laws do not adequately limit how AI could analyze large-scale data for domestic surveillance, potentially undermining constitutional protections.

Amodei said Anthropic hopes the Pentagon reconsiders but would facilitate a smooth transition if the contract is canceled. He also claimed the department threatened to invoke the Defense Production Act to compel removal of safeguards. Undersecretary of Defense Emil Michael criticized Amodei on social media, accusing him of misrepresenting the situation and asserting that the Pentagon would follow the law without yielding to pressure from a private company.

Anthropic, which is backed by Google and Amazon, said it remains open to discussions and committed to supporting the Department of Defense. More than 200 employees from Google and OpenAI signed an open letter supporting Anthropic’s stance, while both companies declined to comment.