Pentagon signs classified AI deals with OpenAI, Google, Nvidia, and others, excluding Anthropic after supply-chain risk designation
The AMW Read
Updates the Anthropic case-study with a major government exclusion and the Pentagon's hyperscaler-distribution pattern, carrying cross-segment geopolitical significance.
Pentagon signs classified AI deals with OpenAI, Google, Nvidia, and others, excluding Anthropic after supply-chain risk designation
The US Department of Defense has announced agreements with OpenAI, Google, Microsoft, Amazon, Nvidia, xAI, and startup Reflection to use their AI tools in classified settings. The deals cover lawful operational use of AI systems, advancing the Pentagon's goal of becoming an AI-first fighting force. Notably, Anthropic was excluded after being declared a supply-chain risk, following a dispute over red lines on mass surveillance and fully autonomous weapons, despite a prior $200 million classified contract. Anthropic is currently under a temporary injunction after suing the government.
Why it matters: This event enshrines the Pentagon's hyperscaler-distribution moat — picking AI labs that accept broad military use while sidelining a top-tier lab (Anthropic) that refused to bend on safety guardrails. It signals that access to defense contracts is now a competitive differentiator, potentially reshaping the frontier-model market by tilting compute, revenue, and talent toward government-compliant labs. The exclusion also underscores the rising importance of supply-chain risk designations as a de facto regulatory lever.
Expert take: This is a structural shift in the AI industry substrate. The Defense Department's move effectively creates a two-tier system: labs willing to endorse any lawful use gain preferential access to massive government contracts and compute resources, while those imposing restrictions face market exclusion. Anthropic's Mythos model, described as a powerful security tool that can find and patch cyber vulnerabilities, remains separately relevant, but the company's principled stance may cost it long-term influence. The pattern echoes earlier acqui-licensing dynamics and reinforces the capital-cycle advantage of hyperscaler-backed labs.
