Pentagon Signs Classified AI Agreements with Seven Firms

According to a Pentagon statement and reporting by Reuters, the Department of Defense reached agreements with seven AI firms, OpenAI, Google, Microsoft, Amazon Web Services, NVIDIA, SpaceX/xAI, and the startup Reflection, to permit use of their AI systems on classified DoD networks. Reuters reports the deals enable integration into the DoD's Impact Level 6 and 7 environments. CNN and Reuters report the Pentagon highlighted its GenAI.mil platform has been used by more than 1.3 million Defense Department personnel.

The coverage notes a conspicuous exclusion: Anthropic. Reuters and CNN report the Pentagon previously designated Anthropic a "supply-chain risk" earlier this year and has not been included in the new agreements. Reporting by The Verge and CNN documents that Anthropic sued the government in response and that a federal judge recently issued a temporary injunction related to the dispute.

Editorial analysis - technical context: Integrating multiple vendor models into classified networks typically requires agreed operational controls for data residency, inference isolation, and security validation. Public reporting does not enumerate the technical controls the Pentagon will require for each provider; Reuters and The Verge note the stated objective of avoiding "vendor lock" by broadening available providers. For practitioners, enabling models to run at Impact Level 6 and 7 implies additional compliance work on model audit trails, hardened inference hosts, and continuous monitoring systems, which are standard requirements for classified operational environments.

The agreements expand the set of vendors who can support classified DoD workflows after a period when Anthropic's model was the primary option inside the classified boundary, a point emphasized by Reuters and CNN. Public reporting frames the moves as part of the Pentagon's effort to diversify suppliers and accelerate the integration of generative AI into planning, logistics, targeting, and other operational tasks. The presence of major cloud and chip vendors such as AWS and NVIDIA, along with platform providers OpenAI and Google, widens the range of inference stacks and deployment patterns available to the DoD, according to multiple outlets including WSJ and Forbes.

Editorial analysis: From a security and procurement perspective, the exclusion of Anthropic after a supply-chain designation creates legal and operational friction; The Verge, CNN and Reuters document litigation and internal resistance from staff who have preferred Anthropic's tools. For practitioners, comparable shifts in vendor authorizations often produce short-term interoperability and tooling headaches as teams adapt CI/CD, model validation tests, and access controls to multiple vendor APIs and on-prem inference endpoints.

• •Whether the Pentagon publishes technical baseline requirements or a standard security framework for vendors operating at Impact Level 6 and 7, as that will determine integration effort for different model architectures.
• •How the Anthropic litigation proceeds and whether federal rulings alter the companys eligibility for future classified work, as reported by The Verge and CNN.
• •Implementation details for GenAI.mil and metrics on operational uptake beyond the 1.3 million user figure cited in Pentagon commentary by Reuters and CNN.
• •How contractors and internal DoD teams manage the transition away from a de facto single-vendor dependency toward a multi-vendor environment, including changes to model evaluation and red-team processes.

Reporting sources include Reuters, CNN, The Verge, The Wall Street Journal, and related coverage in Forbes and The New York Times.