The Department of Defense is currently assessing AI models from OpenAI, Google, and xAI’s Grok as potential substitutes for Anthropic’s Claude. This evaluation follows the decision in late February to categorize Anthropic as a supply-chain risk, declared by Defense Secretary Pete Hegseth. The testing commenced on March 1, 2026, in light of Anthropic’s refusal to relax safety measures that prohibit the use of its AI for mass surveillance or lethal autonomous systems.
In July 2025, Anthropic had secured a considerable contract valued at up to $200 million intended for deploying Claude within classified military networks. However, tensions escalated when Pentagon officials demanded more extensive utilization of the technology, particularly for national security and surveillance operations. This disagreement became widely known in January 2026 when it surfaced that Claude had been involved in intelligence activities focused on Iran. Subsequently, feeling pressured by these developments, Hegseth officially denounced Anthropic’s products as a supply-chain risk, prompting his office to seek alternatives.
A dedicated group of 25 military testers is conducting evaluations through the GenAI.mil platform, which operates separately from the existing Maven Smart System. During this assessment, the military personnel are scrutinizing competing AI models under review from OpenAI, Google, and Grok, aimed at replacing Anthropic’s technology. This situation has led contractors who previously integrated Claude into their systems to seek substitutes within a specified six-month timeframe. Meanwhile, Anthropic has responded by initiating lawsuits concerning the circumstances surrounding the contract.
As the situation unfolds, other tech companies, including OpenAI, are actively pursuing classified contracts with the Department of Defense. Founded to provide a safety-centric alternative to OpenAI, Anthropic's corporate mission hinges on responsible AI development with established restrictions. Failing to reach an agreement with a client on a $200 million contract raises significant questions about the viability of these restrictions in defense settings.