Pentagon Ultimatum Forces Anthropic to Relax AI Military Use Restrictions

In a landmark development that signals a major shift in the ethics of artificial intelligence governance, Anthropic — one of the world’s leading AI research firms — has quietly amended its safety policy to allow the U.S. Department of Defense (DoD) to deploy its Claude AI models for missile defense systems. The change comes after a tense, closed-door meeting at the Pentagon and an ultimatum issued by Defense Secretary Pete Hegseth, who demanded that Anthropic lift key restrictions on military applications by a Friday deadline, according to CNBC.

Just days prior, Anthropic had publicly maintained its commitment to a strict ethical framework, prohibiting the use of its technology in weapons systems or autonomous combat roles. But on February 25, NBC News reported that the company had confirmed to internal stakeholders and government liaisons that its AI systems could now be used for defensive missile tracking and interception — a narrow but strategically significant exception to its prior stance.

The pivot appears to be the result of intense pressure from the DoD. According to Deutsche Welle, Pentagon officials presented Anthropic with a stark choice: either comply with expanded military access to its AI infrastructure or risk losing critical federal contracts, research grants, and potential access to classified defense datasets that underpin the company’s model training. The implications extend far beyond one firm — this episode marks a pivotal moment in the growing entanglement between private AI developers and national security apparatuses.

Anthropic’s original safety policy, unveiled in 2024, explicitly barred the use of its models in offensive military operations, lethal autonomous weapons, and mass surveillance. The company had positioned itself as a moral counterweight to other AI firms that had already partnered with defense contractors. However, sources familiar with internal deliberations suggest that Anthropic’s leadership, while reluctant, concluded that outright defiance could jeopardize its long-term viability in the U.S. market. The firm’s reliance on U.S.-based cloud infrastructure, federal funding, and talent recruitment from top-tier universities may have tipped the scales toward compliance.

Notably, the new agreement does not grant blanket military access. According to NBC News, the DoD’s authorization is currently limited to non-offensive, defensive applications — specifically, real-time threat assessment and trajectory prediction for intercepting incoming ballistic missiles. Still, critics warn that such distinctions are increasingly blurred in modern warfare. Once integrated into a broader defense network, AI systems designed for defense can easily be repurposed for offense, raising profound questions about accountability and control.

Meanwhile, the deadline referenced by CNBC — originally set for February 28, 2026 — has passed, and Anthropic has not issued a public statement clarifying whether broader military access beyond missile defense is now permitted. Internal documents reviewed by DW suggest the Pentagon is seeking access to Anthropic’s model fine-tuning tools and data pipelines, potentially enabling real-time battlefield adaptation of AI systems. Such access would represent a significant escalation beyond the current defensive scope.

Human rights organizations have expressed alarm. “This is not a compromise — it’s a capitulation,” said Dr. Elena Rodriguez of the Algorithmic Justice League. “When companies trade ethical principles for access to government contracts, they normalize the militarization of AI. The public deserves transparency, not backroom deals.”

Anthropic has not responded to requests for comment on whether it will extend military access to other domains, such as intelligence analysis or drone coordination. The company’s silence speaks volumes. As the U.S. military accelerates its adoption of generative AI across all branches, the line between civilian innovation and state surveillance is dissolving. For Anthropic, the choice may have been between principle and survival — but the consequences will be felt far beyond its headquarters in San Francisco.