Pentagon signs classified AI deals with 8 companies, excludes Anthropic over safety dispute
Tags AI · Defense · Policy
The Pentagon formalized classified-network AI agreements with eight companies — SpaceX (including xAI), OpenAI, Google, NVIDIA, Reflection AI, Microsoft, AWS, and Oracle — while explicitly excluding Anthropic. Anthropic was designated a 'supply chain risk' in March 2026 after refusing to allow Claude's use for autonomous weapons and mass surveillance. A California district court issued a preliminary injunction in Anthropic's favor, but the D.C. Circuit Court of Appeals denied an emergency stay, leaving the designation in effect. Despite the exclusion, Anthropic's annualized revenue surged to $44 billion in early May (up from $30 billion weeks prior), approaching a $1 trillion private valuation. The Pentagon's approach signals a preference for a layered AI stack with multiple vendors rather than a single supplier.
Technical significance
The Pentagon's vendor selection establishes that usage restrictions on lethal applications are now a structural disqualifier for defense AI contracts, not a negotiating point. This creates a clear market signal: companies that want DoD revenue at scale face pressure to remove or narrow autonomous weapons restrictions. Anthropic's $44B ARR despite exclusion demonstrates that commercial AI demand can more than compensate for lost defense revenue, but the precedent set — that policy choices about lethal AI use have direct commercial consequences — will shape the entire industry's defense posture.