Nvidia, Palantir and Booz Allen demand data guarantees from Anthropic and OpenAI

Following Anthropic’s June policy change that retains prompts for 30 days and flagged content for up to two years, Nvidia, Palantir and Booz Allen Hamilton have each set conditions on the use of Anthropic and OpenAI models. Palantir demands an irrevocable guarantee that no corporate data will be stored before deploying Claude, while Nvidia limits Anthropic’s models to internal, low-risk projects and routes proprietary work to its own Nemotron models.

Booz Allen has prohibited its employees from running Anthropic’s commercial model on cybersecurity assignments involving client data. The move reflects broader industry anxiety, echoed by Microsoft and other firms, about potential data leakage and training misuse. In response, Nvidia and Palantir launched a joint platform that combines Palantir’s software with Nvidia’s open-source Nemotron models, allowing customers to keep operational data on their own systems.

Anthropic introduced Enterprise Frontier Safeguards, letting businesses store activity data in their own cloud storage under their encryption keys, but it does not meet Palantir’s demand for a non-revocable guarantee. The dispute highlights the growing tension between AI providers and enterprise customers over data privacy and model training practices.