Major U.S. technology and defense companies are restricting or reassessing their use of Anthropic’s closed-source AI models after the company reserved the right in June to retain user usage logs for 30 days to counter sophisticated and novel attacks. Nvidia has limited Anthropic model use to less-sensitive work and directs confidential tasks to its internally deployed Nemotron model, while Booz Allen Hamilton has banned Anthropic’s commercial models for classified cybersecurity work. Palantir is requiring an irrevocable zero data retention commitment before integrating Anthropic models. The concerns have also strengthened interest in locally deployed open-source AI models and allowed Microsoft to promote isolated cloud environments and proprietary AI services. The Information first reported the changes on September 14, with Reuters and other media later reporting on the issue. Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman and xAI head Elon Musk have separately supported slowing frontier AI development, but the dispute highlights a broader trust problem surrounding how closed-source providers handle enterprise data.