Corporate AI security shift: major firms restrict use of OpenAI and Anthropic models

Recently, the enterprise AI market has experienced significant changes as several major clients, including NVIDIA, Palantir Technologies, and Booz Allen Hamilton, have started limiting the use of Anthropic and OpenAI's advanced models in sensitive projects due to concerns over handling proprietary data. The shift was triggered by a policy change at Anthropic earlier this year, which allows its Fable model to retain customer data for up to 30 days to detect misuse. Despite Anthropic's assurances that the data would not be used for training, many clients opted to isolate sensitive information. In response to potential data security risks, companies have adopted various strategies. NVIDIA has restricted the use of the Fable model to lower sensitivity projects, while directing its most confidential business to its own Nemotron model. Booz Allen Hamilton has implemented similar restrictions and requested stricter contractual protections. C Spire has signed agreements with both OpenAI and Anthropic to prohibit the use of its data for training, though existing contracts still permit the collection of technical usage information. Novo Nordisk has taken a cautious approach, using Claude for routine tasks while banning proprietary data input. Microsoft has capitalized on the growing emphasis on data security by promoting private cloud environments as a new market opportunity.