Pentagon approves 4 AI vendors for classified systems

The Pentagon has signed agreements that allow Microsoft, Reflection AI, Amazon and Nvidia products to be used on classified military systems, including Impact Level 6 (secret) and Impact Level 7 (top‑secret). These companies join OpenAI, xAI and Google on the department’s list of vendors approved for “any lawful use.” The Pentagon stated the agreements are intended to prevent vendor lock‑in and ensure long‑term flexibility for the Joint Force.

Pentagon officials described the technologies as tools to help military personnel synthesize data, improve situational awareness and support decision-making in operational environments. Until now, most generative AI use inside the Defense Department has been limited to unclassified tasks such as drafting documents, summarizing text and internal research.

The approvals widen the pool of suppliers available for classified work and reduce the operational effect if an individual company withdraws or changes policies. Defense officials noted that prior disputes between companies and employees over military use of technology have affected availability; Google and Amazon previously disciplined staff who protested their companies’ involvement in defense projects.

The actions followed a public breakdown in relations with Anthropic. The Pentagon canceled a planned $200 million contract with Anthropic after disagreements over the “any lawful use” clause. Anthropic’s CEO Darius Amodei warned the clause could allow government uses the company did not want to support, including domestic surveillance and autonomous weapons. Anthropic has filed a lawsuit seeking damages it says resulted from the contract cancellation and related market effects. The Trump administration designated Anthropic a “supply chain risk,” the first time that label was applied to a U.S.-based company. Some government statements criticized the company’s culture.

Anthropic technologies have not been fully excluded from government programs. Claude had been integrated into classified workflows via Palantir’s Maven toolset, and Anthropic’s Mythos model is under evaluation by multiple organizations for cyber defense capabilities. Officials note that those assessments involve several intelligence and security agencies and some foreign partners, but they have not disclosed the full list of evaluators.

The Pentagon did not clarify whether the expanded approvals cover domestic deployments inside U.S. borders. The White House stated it continues to engage across government and industry, including with AI labs, to protect the country and the American people.