The Pentagon finalized artificial intelligence agreements over the weekend with seven major technology companies, expanding the use of AI inside classified Defense Department environments. The agreements cover OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, and SpaceX, with Reflection AI also reportedly included in a separate track. Anthropic, the AI safety lab whose Claude model has been used by federal agencies under earlier framework contracts, was not included in the new deals.
Defense Department officials cited a contract dispute and what they described as a supply-chain risk concern as the basis for excluding Anthropic. The dispute centered on the company's published acceptable use policy, which prohibits the use of Claude for autonomous weapons systems, mass surveillance, and certain cyber offensive operations. Pentagon negotiators had requested terms that would allow the military to use Claude for "all lawful purposes," language that Anthropic declined to accept. President Donald Trump announced on Truth Social late Friday that the administration would sever the existing federal Anthropic relationship.
The new agreements substantially expand AI access inside classified networks. Defense Department briefing materials describe four primary use categories: intelligence analysis, logistics and supply chain optimization, large-scale data processing, and engineering simulation. Each company has committed to delivering specialized model variants that meet Defense Information Systems Agency security requirements and that can operate in air-gapped environments. The Pentagon estimates the combined value of the agreements at roughly 800 million dollars over four years, with individual ceiling values varying by company.
OpenAI, Google, and Microsoft hold the largest individual contracts under the new framework. Each is expected to deploy specialized versions of frontier models for use by the intelligence community, special operations command, and the four service branches. Amazon Web Services and Oracle will provide cloud infrastructure, while Nvidia will supply specialized hardware including the Blackwell platform. SpaceX is included to provide secure low-latency connectivity through Starshield, the classified version of Starlink. Reflection AI, a smaller research firm, was added to provide alternative model architectures for specific applications.
Anthropic has not publicly responded to the exclusion, though the company's published policies have remained consistent since 2023. The acceptable use policy lists prohibited categories that include developing weapons capable of significant physical harm, conducting mass surveillance without legal authorization, and generating content designed to influence elections. The policy permits use by defense and national security customers within those constraints, and Anthropic had previously delivered Claude through Amazon Web Services to multiple federal agencies under a framework that respected those limits.
Industry analysts have raised concerns about the implications of the exclusion. The decision effectively penalizes a company for declining to remove safety constraints that the company has applied uniformly to all customers, including private sector enterprises. Several Republican lawmakers, including senators on the Armed Services Committee, have asked the Defense Department for additional briefings on the contract structure and the precedent it sets. Democratic lawmakers, including Senator Ron Wyden of Oregon, have requested additional disclosure on what categories of use the seven approved companies have agreed to permit.
The Pentagon agreements come during a period of significant federal AI policy activity. The administration released an AI action plan in March that calls for accelerated military AI adoption and reduced regulatory friction for federal AI procurement. Office of Management and Budget guidance issued in April directs federal agencies to identify AI use cases for procurement consideration on a quarterly basis. The Defense Department alone has identified more than 600 candidate use cases.
Anthropic recently disclosed quarterly results that placed annualized revenue at 19 billion dollars, up from 5.4 billion at the start of the year. The company serves more than 8,140 enterprise customers including 43 of the top 50 law firms and 17 of the top 20 health systems. Google announced in late April that it would invest up to 40 billion dollars in Anthropic in cash and compute. The company has not announced an IPO timeline but has signaled that public market consideration is being explored.
The Defense Department contracts are subject to standard federal acquisition review and could face protest from non-selected vendors. A formal Government Accountability Office bid protest window remains open through mid-May. Anthropic has not signaled whether it intends to file a protest, though defense law specialists note that the unusual circumstances around the exclusion could provide grounds for review.
For technology workers and federal contractors in Tennessee, the agreements create both opportunities and questions. Oak Ridge National Laboratory, which conducts significant AI research under Department of Energy supervision, is expected to be a major implementation site for several of the new deals. Vanderbilt University Medical Center, which holds research relationships with multiple AI companies, has not commented on the policy implications. Senator Marsha Blackburn and Senator Bill Hagerty have not issued statements on the exclusion. Representative Steve Cohen, who represents Memphis, called for additional Congressional oversight of military AI procurement in a statement issued Sunday evening.
