Anthropic’s relationship with the U.S. Department of Defense broke down sharply in February 2026. According to a major TIME report, the AI company was designated a “national security supply chain risk” by the Trump administration after it refused to loosen limits on how Claude could be used. The U.S. government then moved to stop using its software, while OpenAI stepped in to take over the military work.
Two usage limits became the center of the dispute
The clash focused on two boundaries pushed by Anthropic CEO Dario Amodei: Claude should not be used for fully autonomous weapons systems, and it should not be used for mass surveillance of U.S. citizens. Pentagon officials wanted contract language broad enough to allow “all lawful use.” Anthropic would not agree. Defense officials argued that the company’s insistence on guardrails and its repeated debates over hypothetical scenarios weakened the partnership.
On February 27, the Trump administration formally labeled Anthropic a national security supply chain risk. TIME described it as the first known case of a U.S. domestic company receiving that designation. Defense Secretary Pete Hegseth then said companies doing business with the government could no longer work with Anthropic. On March 9, Anthropic sued the U.S. government in an effort to overturn the blacklist decision.
Claude had already become embedded in U.S. military workflows
Before the rupture, Claude had become one of the AI systems most relied upon by the U.S. government. The report says it was the first frontier AI model approved for deployment in classified environments. In January 2026, Claude was also used in planning and intelligence analysis tied to the operation that led to the arrest of Venezuelan president Nicolás Maduro. The exact extent of its role remains unclear, but the episode was presented as one of the earliest cases of frontier AI being deeply involved in a real military mission.
That same period saw contract talks between Anthropic and the Pentagon become increasingly tense. U.S. officials said they were concerned the company might restrict use of the model at a critical moment. Anthropic denied trying to control military operations on a case-by-case basis.
Rapid growth collided with a safety-first identity
The confrontation came as Anthropic was rising to the top tier of the AI industry. TIME reported that the company had recently raised $30 billion and reached a $380 billion valuation, putting it ahead of legacy giants such as Goldman Sachs, McDonald’s, and Coca-Cola. Claude and Claude Code were key drivers of that rise. The coding agent product alone had crossed $1 billion in annualized revenue by the end of 2025, then climbed to $2.5 billion by February 2026.
At the same time, Anthropic continued to frame itself around safety. In February 2025, a controlled test suggested a new Claude release might help terrorists synthesize biological weapons. The company delayed the launch of Claude 3.7 Sonnet by 10 days while teams reviewed the risk. Inside Anthropic, fast-improving model capability was treated as inseparable from biosecurity, cyberattack, and loss-of-control concerns.
Anthropic also eased part of its own safety framework
TIME said Anthropic had long pointed to its Responsible Scaling Policy as evidence that it would slow down if safety measures were not strong enough. The original policy included a binding commitment to pause development under certain conditions. In late February 2026, the company revised that policy and removed the explicit pledge to halt development.
Co-founder and chief scientist Jared Kaplan told TIME that it was unrealistic for Anthropic to keep making unilateral commitments if competitors were pushing ahead at full speed. The revised policy shifted toward transparency, more disclosure around safety testing, and a promise to match or exceed peers on safety investment. For a company built around a safety-first image, that change showed how competitive pressure was already narrowing the room it had to hold the line.