Pentagon AI Contracts Show Deep Integration With Tech Giants

Contracts reveal unprecedented military-tech collaboration

The U.S. Department of Defense signed contracts worth up to $200 million each with four major AI companies in July 2025, according to documents obtained through a Freedom of Information Act lawsuit. The agreements with Anthropic, Google, OpenAI, and xAI committed these firms to building AI prototypes designed to "improve military advantage, military utility, or enhance military decision making" across all branches of the armed forces.

The Intercept obtained more than 400 pages of contract materials through litigation conducted with Legal Advocates for Safe Science and Technology. The documents, many heavily redacted, detail applications ranging from warfighting and automated decision-making to logistics and military intelligence. These prototypes now form the foundation of AI tools deployed on the Pentagon's classified networks.

Two-way information exchange embedded in agreements

The contracts establish far more than a simple vendor relationship. Information-sharing provisions create bidirectional flows of data and expertise between Silicon Valley and the military. The companies agreed to provide access to their most advanced large language models while also committing to "inform DOD and CDAO strategy on frontier AI," according to language in Google's agreement.

Under these terms, AI labs received approval to access sensitive military information, potentially including operational mission briefings, threat assessments, and Pentagon adoption strategies. In return, the companies pledged to offer feedback on defense AI strategies and provide briefings on their own corporate developments, including model performance projections and future capability trends.

Google specifically agreed to brief the Pentagon on "AI tactics and techniques of adversaries" in both classified and unclassified settings. All four contractors committed to conducting training seminars and tabletop exercises with military personnel.

Risk forecasting raises conflict-of-interest concerns

The contracts require AI companies to provide "risk forecasting and threat ideation exercises" — essentially predicting dangers their own products might pose. Heidy Khlaaf, chief scientist at the AI Now Institute and former OpenAI safety engineer, told The Intercept this arrangement creates troubling conflicts of interest.

Khlaaf pointed to recent incidents where semi-autonomous large language models broke into other companies' computer networks during testing. She characterized company-provided risk forecasts as evidence-light and skewed toward "self-beneficial" concerns like speculative existential threats rather than concrete harms. Independent assessment would better serve the public than allowing companies to self-report risks with "life-or-death consequences," she said.

Anthropic dispute centered on classified deployment terms

The relationship between Anthropic and the Pentagon deteriorated during negotiations over follow-up agreements. Anthropic refused to sign amendments allowing classified network deployment without contractual prohibitions against domestic spying and autonomous weapons. Defense Secretary Pete Hegseth designated the firm a "supply chain risk" in March and banned its services from government use, though a federal judge later overturned that decision.

The Pentagon quickly finalized classified deployment deals with Google, OpenAI, and xAI. An amended OpenAI contract dated February 27 includes provisions for embedding company engineers within military units, noting these workers "can be deployed to warfighting support settings including, but not limited to, combatant commands, service components, and theatre components."

Why it matters

These contracts illuminate how deeply AI companies have integrated with military operations despite earlier public commitments against weaponization. OpenAI banned "military and warfare" use in its terms of service until two years ago. Google maintained similar restrictions on surveillance applications. Both companies now actively pursue Pentagon contracts while the actual constraints on their technology's military use remain unclear. The Wall Street Journal reported in March that U.S. Central Command used Anthropic's services for "target identification" during Iran bombardment operations, even during the period when the company was officially banned. When asked if his company's tools were used in a strike that killed 120 Iranian schoolchildren, Anthropic CEO Dario Amodei said he did not know.

Details were first reported by The Intercept based on documents obtained through FOIA litigation.