Pentagon Tests Rival AI Models in Race to Replace Anthropic

The Pentagon has launched trials comparing rival AI models to Anthropic PBC’s Claude, seeking replacements for the system after designating it a supply-chain risk. Tests began March 1, just three days after Defense Secretary Pete Hegseth announced the decision, citing guardrails imposed by Anthropic as a concern. The department is consulting 25 of its ‘power users’ to determine which alternatives best meet military needs. Anthropic’s classification as a supply-chain risk stems from restrictions the company enforces on its AI technology, which the Pentagon views as limiting operational flexibility. The move aligns with broader U.S. efforts to reduce reliance on foreign or commercially constrained AI providers. Hegseth’s directive followed a review of potential vulnerabilities tied to Anthropic’s guardrails, which could restrict military applications. Anthropic is actively contesting the Pentagon’s decision in court, arguing that the designation could result in billions in lost revenue. The company maintains that its guardrails enhance safety and alignment with ethical standards, countering claims that they hinder functionality. Legal proceedings remain unresolved as the Pentagon continues evaluating competing AI systems. The testing phase involves direct comparisons between Claude and other models, focusing on performance, adaptability, and compliance with military requirements. Officials have not disclosed the names of rival AI providers under consideration, emphasizing the competitive and sensitive nature of the evaluation. The outcome could reshape the U.S. defense sector’s approach to AI integration and procurement. If successful, the Pentagon’s shift away from Anthropic could set a precedent for other government agencies to reassess their AI partnerships. The case also underscores growing tensions between commercial AI ethics and military operational demands. As trials progress, the defense department will weigh technical capabilities against broader strategic and security implications.