A dispute between Anthropic and the U.S. Defense Department has led federal agencies to phase out the company’s artificial intelligence systems. Agencies have up to six months to transition away from Anthropic’s technology.
The Pentagon also designated the firm a “Supply-Chain Risk to National Security” after negotiations over revised contract terms ended without agreement.
The disagreement focused on how Anthropic’s AI models could be used in classified military and intelligence environments.
Anthropic sought written limits preventing the use of its systems for mass domestic surveillance and for fully autonomous weapons operating without human oversight. The Defense Department requested authorization to use the technology for all lawful military purposes.
During the same period, OpenAI finalized a separate agreement with the Pentagon that allows classified deployments under defined contractual and technical controls.
Why It Matters: The outcome affects how AI companies gain and retain access to classified federal work. Decisions about permissible use shape product design, deployment architecture, legal exposure, and procurement standards, while also influencing how much operational discretion the government expects in future AI partnerships.
- Anthropic’s Contract Breakdown and Federal Phase-Out: Anthropic previously held a $200 million Pentagon contract that included classified applications tied to military intelligence operations. The company sought explicit contractual language limiting certain uses of its AI. When the Defense Department did not agree to those terms by its stated deadline, the government directed agencies to stop using the company’s systems and applied a supply-chain risk designation. Anthropic has said it plans to challenge that decision in court.
- OpenAI’s Classified Agreement Structure: OpenAI reached an agreement permitting the use of its AI in classified environments. The company states that its contract includes references to existing U.S. law and Department of Defense directives governing surveillance and autonomous weapons. OpenAI also said its deployment will remain cloud-based, that it will retain control over its safety systems, and that cleared personnel will participate in oversight of how the models are used.
- Different Approaches to Enforcing Guardrails: Anthropic emphasized upfront contractual restrictions that define prohibited uses in advance. OpenAI described layered safeguards that combine contract language, deployment architecture, and continued company involvement in system oversight. The difference centers on how compliance is enforced, either through stricter pre-negotiated limits written into the agreement or through technical controls and monitoring embedded within the deployment model.
- Supply-Chain Risk Designation and Legal Review: The designation of Anthropic as a supply-chain risk is uncommon for a U.S.-based AI developer. Such classifications can restrict eligibility for future contracts across agencies. A legal challenge could clarify how procurement authorities apply supply-chain standards when disputes concern usage governance instead of ownership structure or cybersecurity vulnerabilities.
- Implications for Future Defense AI Contracts: The differing outcomes for Anthropic and OpenAI may influence how AI firms negotiate with federal agencies going forward. Companies may place greater emphasis on how their systems are deployed, how compliance is verified, and how responsibilities are divided between vendor and government users. The case illustrates that governance terms are becoming central to whether advanced AI tools are approved for classified defense work.
Go Deeper -> AI executive Dario Amodei on the red lines Anthropic would not cross – CBS News
Our agreement with the Department of War – OpenAI
Trump admin blacklists Anthropic as AI firm refuses Pentagon demands – CNBC
Trusted insights for technology leaders
Our readers are CIOs, CTOs, and senior IT executives who rely on The National CIO Review for smart, curated takes on the trends shaping the enterprise, from GenAI to cybersecurity and beyond.
Subscribe to our 4x a week newsletter to keep up with the insights that matter.
