Pentagon signs classified-network AI deals with seven vendors, excludes Anthropic

TL;DR:

• The US Department of Defense has signed agreements with SpaceX, OpenAI, Google, NVIDIA, Reflection AI, Microsoft and Amazon Web Services to deploy their AI on classified secret and top-secret networks.
• Anthropic is excluded; the Pentagon designated it a supply-chain risk in March and the two sides are in litigation, though President Trump said last week the firm was “shaping up”.
• The Pentagon’s main AI platform GenAI.mil has reached over 1.3 million Defense Department personnel in five months; CTO Emil Michael told CNBC Anthropic remains a supply-chain risk but the firm’s Mythos cyber-capable model is a “separate national security moment”.

The vendor list, announced on Friday, formalises a faster onboarding process the Pentagon adopted after the Anthropic dispute escalated. New AI entrants now report being incorporated onto secret-level data environments in under three months, down from 18 months or longer. Reflection AI, the lesser-known name on the list, raised $2bn in October and is backed by 1789 Capital, where Donald Trump Jr. is a partner.

Context

The deal architecture explicitly targets “vendor lock” — Pentagon language that Reuters reads as a reference to overdependence on Anthropic, whose tools military staff and IT contractors continue to view as superior despite orders to remove them within six months. Pentagon staffers’ resistance to dropping Anthropic has been reported separately by Reuters; the new multi-vendor list is the procurement-side response.

For UK Ministry of Defence and DSIT procurement watchers, the read-across is direct. The UK has not designated any frontier-AI vendor a supply-chain risk to date, and the Defence AI Strategy refresh issued in 2025 commits to multi-vendor frameworks for similar reasons. The Pentagon’s ability to compress its accreditation timeline from 18 months to three months is the kind of operational reform UK defence buyers have publicly aspired to but not yet matched. The Anthropic dispute also highlights a regulatory exposure the UK MoD will face: how to handle a vendor whose product is judged commercially superior but politically or contractually unworkable.

The Mythos point is the more subtle signal. Defense CTO Emil Michael’s framing of Mythos’s cyber capabilities as “a separate national security moment” — distinct from the supply-chain dispute — suggests the Pentagon is treating frontier offensive-cyber AI capability as a category of its own, with its own access regime. UK NCSC’s recent warning about AI accelerating vulnerability discovery sits in the same conceptual frame, even though the policy mechanisms differ.

Looking forward

Anthropic’s path back into the Pentagon hinges on the Trump administration’s evolving posture; the President’s “shaping up” remark suggests negotiation rather than permanent exclusion. For UK buyers, the practical takeaway is to verify whether their incumbent foundation-model vendors’ US classifications cascade into UK eligibility decisions for sensitive work, and whether the UK MoD’s own list mirrors or diverges from this one.