Pentagon broadens AI defence suppliers as Anthropic dispute drags on

TL;DR:

• The Pentagon has signed agreements with Microsoft, Reflection AI, Amazon and Nvidia, joining OpenAI, xAI and Google as suppliers cleared for classified deployment “for any lawful use” at Impact Levels 6 and 7.
• Anthropic remains designated a “supply chain risk” — the first US-based company ever given that label — after it refused to allow its technology for civilian surveillance or autonomous weapons; the company is suing over a cancelled $200 million contract.
• Resultsense view: behind the public dispute, Anthropic’s Mythos cyber-defence model is reportedly under assessment by 40 organisations including the US National Security Agency and, according to AI News, possibly the UK’s MI5 — meaning the practical procurement picture is more nuanced than the political one.

The Department of Defense’s expansion of its approved AI vendor list is framed in the Pentagon’s own statement as a deliberate move to “prevent AI vendor lock-in and ensure long-term flexibility for the Joint force”. The four new entrants will support classified data work at the highest impact levels, and contribute to what the Pentagon describes as an “AI-first fighting force”.

What the new agreements cover

Current generative AI use inside the DoD has been concentrated in non-classified document drafting, summarisation and research. The new suppliers’ tools will extend into classified data synthesis, situational understanding, and warfighter decision support. The Pentagon statement does not explicitly address whether those use cases include domestic deployments inside US borders.

By broadening the supplier base, the department insulates itself from any individual vendor’s policy positions — a pointed reference to past episodes at Google and Amazon, where staff protests resulted in firings over military contracts.

The Anthropic standoff

Anthropic CEO Dario Amodei publicly objected to the “any lawful use” framing, arguing it would permit US government deployment of Anthropic technology for civilian surveillance and autonomous weapons — areas the company says it walls off. The Pentagon cancelled a $200 million contract; Anthropic took the matter to court. White House sources have separately described the company as “woke”, and the supply-chain risk designation is unprecedented for a US-headquartered AI firm.

According to Axios, the administration may be looking for a path back. A White House source described efforts to “save face and bring ‘em back in”, and Anthropic’s Claude coding model is reportedly still in active use by US government security organisations throughout the dispute. Mythos, Anthropic’s cyber-capability model, is being assessed by 40 organisations worldwide; only 12 have been named publicly.

Looking forward

For UK readers, the practical signal is not the headline dispute but the procurement reality: the US Pentagon is hardening its AI supplier base while the UK builds its own frontier-AI accountability framework around the AI Safety Institute and ICO. Whether Anthropic’s UK relationships — including the reported MI5 Mythos assessment — survive the political turbulence in Washington will be a live question for British security buyers over the next quarter.