Major companies restrict use of Anthropic and OpenAI models over data handling concerns

Several major companies, including NVIDIA, Palantir Technologies, and Booz Allen Hamilton, are limiting their use of Anthropic and OpenAI's cutting-edge AI models due to concerns over how proprietary data is handled. The restrictions stem from a policy change by Anthropic in June, which allowed its Fable model to retain customer data for 30 days to detect misuse. While Anthropic claims the data is never used for model training, some companies prefer to isolate sensitive information. NVIDIA has restricted Fable to less sensitive projects like open-source software development, while using its own Nemotron model for confidential operations. Booz Allen Hamilton has also implemented similar measures with stricter contractual protections. C Spire has signed agreements with both OpenAI and Anthropic to prevent data training, though the contracts still permit the collection of technical usage data. Novo Nordisk has taken a more cautious approach, continuing to use Claude for certain tasks while banning proprietary data input. Microsoft is capitalizing on this trend by promoting a private cloud environment for AI models, which operates on managed servers and does not transmit user data externally, despite the high costs involved.