Anthropic secures $517 billion Pentagon computing contracts despite public breakup

Newly revealed documents show that in July 2025, the Pentagon signed $200 million AI prototyping contracts with Anthropic, Google, OpenAI, and xAI. Although Anthropic was banned as a 'supply chain risk' in March 2026 by Secretary Pete Hegseth after refusing to deploy on classified networks without safety guardrails, a federal judge overturned the ban in August 2026. Meanwhile, OpenAI finalized a rushed deal in February 2026, embedding its engineers directly into military combatant commands.

Pentagon AI prototype contracts

• ▪The artificial intelligence prototypes developed under the July 2025 Department of Defense contracts formed the basis of AI tools deployed across the Department of Defense's classified networks.
• ▪The artificial intelligence prototypes developed under the July 2025 Department of Defense contracts were designed to improve military advantage, utility, and decision-making across the armed forces.
• ▪In July 2025, the U.S. Department of Defense signed contracts worth up to $200 million each with Anthropic, Google, OpenAI, and xAI to build prototypes of artificial intelligence tools.

Information sharing provisions

• ▪Under the July 2025 contracts, the artificial intelligence companies were approved to receive sensitive military information, including benchmark datasets, operational mission briefings, and future AI adoption plans.
• ▪Under its July 2025 contract, Google agreed to provide the Pentagon with briefings on the artificial intelligence tactics and techniques of United States adversaries in both classified and unclassified settings.
• ▪The July 2025 Department of Defense contracts with Anthropic, Google, OpenAI, and xAI included an 'information sharing' provision establishing a two-way exchange of data and expertise.

Anthropic supply chain ban

• ▪Despite the March 2026 ban by Secretary of Defense Pete Hegseth, the U.S. Central Command used Anthropic's large language model hours after the ban was announced, according to a Wall Street Journal report.
• ▪In August 2026, a federal judge overturned the Department of Defense's ban on Anthropic's services, ruling the ban illegal.
• ▪Anthropic refused to sign a follow-up contract for deploying its technology on classified military networks without contractual prohibitions against domestic spying and autonomous weapons.
• ▪In March 2026, Secretary of Defense Pete Hegseth designated Anthropic a 'supply chain risk' and banned its services from government use after Anthropic refused to sign a follow-up deployment contract.

OpenAI classified network deployment

• ▪On February 27, 2026, OpenAI finalized a follow-up agreement with the Department of Defense to deploy its artificial intelligence tools on classified military networks.
• ▪The February 27, 2026, contract amendment allowed OpenAI engineers to be deployed to warfighting support settings, including combatant, service, and theatre components.
• ▪OpenAI's February 27, 2026, contract amendment included a 'System Oversight' section that was redacted in its entirety, unlike equivalent agreements for xAI and Google.
• ▪An amended contract dated February 27, 2026, required OpenAI to embed company engineers within the military to assist U.S. government employees in implementing and utilizing its AI system.

AI risk assessment concerns

• ▪Heidy Khlaaf, chief scientist at the AI Now Institute, criticized the self-reporting risk provisions, stating that trusting companies to self-report risks constitutes a conflict of interest and subverts democratic processes.
• ▪The July 2025 contracts required the artificial intelligence companies to provide the Pentagon with risk forecasting and threat ideation exercises regarding their own products.
• ▪Sophia Goodfriend, a University of Cambridge research fellow, stated that the contracts cast doubt on the contractual limits AI labs intended to place on the Department of Defense's use of their systems.

Tech company military training

• ▪The July 2025 contracts required engineers and policy experts from Anthropic, Google, OpenAI, and xAI to conduct gamified tabletop simulation exercises with the Pentagon.
• ▪Under its July 2025 contract, Google agreed to lead multiple seminars for the military on the responsible deployment of artificial intelligence.
• ▪Under the July 2025 contracts, the artificial intelligence companies agreed to provide training, educational seminars, and presentations to the military at technical exchange meetings.

Debatable claims

• ▪OpenAI should not allow its engineers to be deployed to military warfighting settings
• ▪The Pentagon should replace AI companies' self-reported risk forecasts with independent assessments
• ▪Anthropic was justified in refusing the Pentagon's classified network deployment contract