The U.S. Department of Defense is locked in a dispute with artificial intelligence startup Anthropic over safeguards that would limit how the military and intelligence agencies can deploy the company’s AI technology, including its use in autonomous weapons targeting and domestic surveillance, Reuters reported, citing people familiar with the matter.
According to Reuters, the disagreement has emerged during negotiations tied to a Pentagon contract worth up to $200 million and represents an early test of whether major AI developers can influence how Washington deploys increasingly powerful artificial intelligence systems in national security operations.
After months of discussions, talks between Anthropic and the Pentagon have stalled, six sources told Reuters on condition of anonymity. At the center of the impasse are Anthropic’s internal rules governing how its AI models can be used, which Pentagon officials argue should not restrict government deployment as long as U.S. law is followed.
A spokesperson for the Defense Department, which the Trump administration has renamed the Department of War, did not immediately respond to requests for comment. Anthropic said in a statement that its AI systems are “extensively used for national security missions by the U.S. government” and that it remains in “productive discussions” with the department.
The dispute could pose a risk to Anthropic’s growing Pentagon business at a sensitive moment for the San Francisco-based startup. The company is preparing for a potential public offering and has invested heavily in expanding its national security work while seeking to shape U.S. government AI policy.
Anthropic was among several major AI developers awarded Pentagon contracts last year, alongside companies including Google, OpenAI and Elon Musk’s xAI.
Sources told Reuters that Anthropic executives raised concerns in talks with U.S. officials that its AI tools could be used to conduct surveillance on Americans or assist in weapons targeting without adequate human oversight. Pentagon officials, citing a January 9 memo outlining AI strategy, have pushed back, saying commercial AI tools should be deployable regardless of company policies if their use complies with the law.
Despite the standoff, Reuters noted that the Pentagon may still need Anthropic’s cooperation. The company’s models are designed to resist actions that could cause harm, and Anthropic engineers would likely be required to modify the systems for military use.
The tensions reflect broader unease in Silicon Valley about how government agencies may use advanced AI technologies. Anthropic CEO Dario Amodei recently wrote that AI should support national defense “in all ways except those which would make us more like our autocratic adversaries,” underscoring the philosophical divide now playing out between the company and U.S. defense officials.
The outcome of the dispute could set an important precedent for how the U.S. military works with private AI developers as artificial intelligence becomes increasingly central to modern warfare and intelligence operations.
