A dispute between the U.S. military and one of the world’s leading AI firms has exposed a growing divide over how artificial intelligence should be used in warfare and surveillance. U.S. Defense Secretary Pete Hegseth has given Dario Amodei, chief executive of Anthropic, a firm deadline to loosen safety limits built into the company’s AI model. If the company refuses, officials say the Pentagon may cancel a $200 million contract and take further action that could damage Anthropic’s position in the defense market. At the center of the dispute are the guardrails Anthropic placed on its AI systems. The Pentagon wants broader access so military users can deploy the technology for “all lawful use,” according to people familiar with the negotiations.
Anthropic, however, has drawn firm red lines around two areas: AI-controlled weapons and large-scale domestic surveillance.
Company leaders argue that current AI systems are not reliable enough to control weapons safely. They also say there is no clear legal framework governing mass surveillance powered by advanced AI tools. Because of those risks, Anthropic has refused to remove certain safeguards from its models.
Pentagon Threatens Emergency Powers in Clash with Anthropic
Pentagon officials see the issue differently. One official said legality rests with the military as the end user, not with the technology provider. “You can’t lead tactical operations by exception,” the official told reporters, stressing that the Defense Department operates within existing law.
The disagreement has escalated quickly. Officials warned that if Anthropic does not comply by Friday evening, the government could terminate the contract and invoke emergency powers under federal law to compel cooperation.
The department is also considering labelling Anthropic a “supply chain risk,” a designation that could block defense contractors from using its technology.
Legal experts question how those steps could work together. Katie Sweeten, a former Justice Department liaison to the Defense Department, said the approach appears contradictory. If a company poses a supply chain risk, she argued, it would make little sense to force the same company to provide technology for military use. She suggested the designation could be punitive rather than rooted in security concerns.
Despite the tension, participants described recent meetings as calm and professional. Hegseth reportedly praised Anthropic’s technology and expressed interest in continuing cooperation. Amodei, in turn, thanked defense officials for their engagement but repeated that the company would not cross its safety boundaries.
A High-Stakes Standoff Over National Security and AI Ethics
Anthropic said discussions remain constructive. In a statement released to CNN, the company called the negotiations a “good-faith” attempt to strike a balance between national security requirements and the responsible use of AI. The company pointed out that it already partners with government agencies and was one of the first AI developers to put its models on classified networks.
The impasse may also change the landscape of competition for AI companies vying for defence contracts. A Pentagon official has confirmed that xAI, the AI company founded by Elon Musk, has expressed a willingness to work in a classified environment. Other companies are also getting closer to defence partnerships, possibly filling the void left by Anthropic.
Anthropic has always marketed itself as a safety-conscious alternative in the rapidly evolving AI sector. The firm was established by ex-researchers at OpenAI who left the company over disagreements about the pace of development and safety measures. The firm’s management has stated that strict controls are required as AI systems become increasingly powerful and influential.
The resolution of the conflict could establish an important precedent. Governments are increasingly recognising the strategic value of advanced AI, while AI developers are concerned about potential misuse and a lack of clear guidelines.
A possible forced compliance by the Pentagon could indicate that the interests of national security take precedence over corporate safety policies. On the other hand, Anthropic’s refusal to budge could consolidate the notion that AI companies have the ability to establish their own ethical frameworks even in the face of government pressure.
At the moment, both parties are engaged in ongoing negotiations with a deadline that could impact the future of military AI collaborations.
