Pentagon Picks Eight AI Vendors and Cuts Anthropic Out

The US Department of Defense announced on 1 May 2026 that it has finalised artificial-intelligence agreements with eight technology companies — OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX and Reflection AI. The list is notable for who is on it. It is more notable for who is not.

Anthropic, the maker of Claude and one of the three frontier-model labs by capability, has been excluded. The Pentagon cited a "supply-chain risk" framing, but the underlying story is a dispute over use cases. According to administration officials, Anthropic refused to back down on contractual language that would have allowed the military to use Claude for "all lawful purposes," including autonomous weapons and mass surveillance.

How we got here

President Donald Trump publicly announced earlier this year that the administration would sever ties with Anthropic. The company sued the federal government in response. A federal judge in California blocked the government's attempt to terminate Anthropic's existing contracts last month. The Pentagon's new vendor list is the workaround: the eight selected companies receive expansive use of advanced AI on classified military networks; Anthropic does not.

Anthropic's parallel play

Anthropic has not been passive in Washington. CEO Dario Amodei visited the White House in April for a meeting with Chief of Staff Susie Wiles, days after the company unveiled Mythos — an AI tool aimed at identifying cybersecurity threats and one of the few Anthropic products explicitly designed for government and large-enterprise security customers. The pitch is straightforward: Anthropic can be useful to the US national-security apparatus on defensive cyber even if it will not provide models for offensive weapon systems.

What the eight vendors get

Contract values were not all disclosed individually. Public statements describe expansive access to classified networks, integration with Defense Department platforms, and tooling for analytical, logistical, cyber and — for some vendors — operational use cases. SpaceX's inclusion alongside OpenAI, Google and Microsoft is striking; Reflection AI, a younger entrant, signals the Pentagon's appetite for emerging capability rather than only incumbent Big Tech.

Why this matters beyond Washington

It is the clearest US-government test yet of whether frontier-model labs can hold a usage-policy line against a sovereign customer. Anthropic has — at meaningful commercial cost. The implication for European procurement is non-trivial: the EU's 2026 AI Act enforcement environment converges, in places, with Anthropic's stated red lines. A lab that has demonstrated it will walk away from a contract on principle is, by that demonstration, easier for a European public buyer to trust on alignment claims. Whether that translates into transatlantic contracts at scale is the next question.