• Bipartisan senators intervene with a private letter to resolve the escalating conflict over Anthropic's Claude AI for military applications.
  • The Pentagon has set a Friday 5 p.m. deadline for Anthropic to allow unrestricted lawful use of Claude, threatening contract termination or invocation of the Defense Production Act.
  • Anthropic rejects the offer, citing inadequate safeguards against mass surveillance and fully autonomous weapons, risking its $200 million Pentagon contract.

In a high-stakes standoff over military AI ethics, senators from the Senate Armed Services and Appropriations Committees sent a private bipartisan letter to Anthropic and the Pentagon on February 27, 2026, urging resolution of their dispute. The move comes as tensions escalate, with the Pentagon imposing a Friday 5 p.m. deadline for Anthropic to comply with demands for unrestricted lawful use of its Claude AI in classified environments, or face potential contract loss, supply chain risk labeling, or invocation of the Defense Production Act.

According to people familiar with the matter, the letter was signed by Armed Services Chair Roger Wicker (R-MS), Ranking Jack Reed (D-RI), Appropriations Mitch McConnell (R-KY), and Ranking Chris Coons (D-DE), signaling congressional oversight push amid the impasse. The Pentagon sent a "best and final offer" on Wednesday, with Defense Secretary Pete Hegseth insisting on full access for lawful activities, while Anthropic views the language as allowing circumvention of its restrictions on surveillance and autonomous targeting.

Anthropic, an AI safety-focused startup with a $200 million Pentagon contract awarded in July for national security AI development, has rejected the offer. CEO Dario Amodei publicly defended the company's stance, citing inadequate safeguards against mass surveillance of Americans and fully autonomous weapons, and stated willingness for a smooth transition to alternatives without disrupting military operations. "We're committed to safe AI deployment, and the Pentagon's current approach risks undermining that," Amodei said in a brief statement, though attempts to reach him for further comment were unsuccessful.

Negotiations have been ongoing for months, with the Pentagon meeting Amodei on Tuesday to press its case. The dispute centers on executive versus private sector control over military AI ethics, with the Pentagon arguing surveillance requests are illegal and seeks lawful use only, while Anthropic demands binding restrictions. Industry experts note this reflects a broader shift toward AI safety guardrails versus unrestricted government use, highlighting tensions in defense procurement. Without a deal, the company would be forced into bankruptcy or offboarding from Pentagon projects, potentially disrupting U.S. defense tech supply chains.

Efforts to restructure the agreement have hit a snag, with sources indicating the Pentagon is considering labeling Anthropic a supply chain risk if it fails to comply. The short-term outlook hinges on the Friday evening deadline; non-compliance could prompt congressional legislation on military AI doctrines for human control and accountability. Analysts warn of procurement coercion precedents impacting AI vendor-government ties, as similar cases like Google's Project Maven withdrawal in 2018 signal sector-wide ethics trends. In the meantime, warfighters face potential AI access loss, sparking debate among civil liberties groups over surveillance risks despite the Pentagon's legality claims.