Body
Defense Secretary Pete Hegseth pressed Anthropic’s chief executive, Dario Amodei, during a Tuesday meeting that led to a Friday deadline to broaden military access to the company’s artificial intelligence technology, or face the prospect of losing a Pentagon contract, an AP source said.
The AP report said the meeting was cordial but that Amodei did not change two restrictions he has said Anthropic will not cross: allowing fully autonomous military targeting and enabling domestic surveillance of U.S. citizens. Anthropic, which makes the Claude chatbot, is described as the last of its AI peers to refrain from providing its technology to a new U.S. military internal network.
The dispute centers on what the Defense Department wants to do with AI systems once they are available inside government networks. The AP source and a senior Pentagon official, who both spoke on condition of anonymity, said Pentagon officials warned they could designate Anthropic as a supply chain risk or use the Defense Production Act to give the military more authority to use Anthropic’s products even if Anthropic did not approve of how they would be used.
Anthropic’s approach to limiting how its models can be used has been part of its public positioning. AP said CEO Amodei has previously argued that allowing AI to operate without guardrails could be dangerous, pointing in particular to fully autonomous armed drones and to AI-assisted mass surveillance that could track dissent.
As the Pentagon accelerates adoption of AI in national-security settings, the episode also reflects broader concerns about lethal-force use and access to sensitive information. The AP report said the development, which it noted was also reported earlier by Axios, highlights the tension between developer-set ethical restrictions and the Pentagon’s operational desire for tools that do not come with built-in limitations.
The Pentagon announced last summer it had awarded defense contracts—each worth up to $200 million—to four AI companies: Anthropic, Google, OpenAI, and xAI, the AP report said. It said Anthropic was the first of the companies to be approved for classified military networks, where it works with partners including Palantir, and that xAI’s Grok chatbot says it is ready for classified use.
The report said Hegseth has publicly argued that the U.S. military should have AI systems that can be used without “ideological constraints” that limit lawful applications. In a January speech at SpaceX in South Texas, he said he was shrugging off any AI models “that won’t allow you to fight wars,” and AP reported that he also said the Pentagon’s AI “will not be woke.”
In a statement after the Tuesday meeting, Anthropic said it “continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s national security mission in line with what our models can reliably and responsibly do,” AP reported. The report also said analysts described Anthropic’s situation as a test of whether its safety-focused posture can be reconciled with the Defense Department’s policy expectations.
Owen Daniels, an associate director of analysis and fellow at Georgetown University’s Center for Security and Emerging Technology, said Anthropic’s leverage in the bargaining was limited because other AI companies had been willing to comply with the department’s policy on lawful applications. Separately, Amos Toh, senior counsel at the Brennan Center’s Liberty and National Security Program at New York University, said the “breakneck” adoption of AI shows the need for oversight or regulation by Congress, and he said the law is not keeping up with how quickly the technology is evolving.
The AP report also said Anthropic has previously found itself at odds with the Trump administration on AI policy and regulation, even as it has remained a partner to other companies in the tech ecosystem.