Palantir, Nvidia and Booz Allen are curbing employee use of Anthropic and OpenAI’s AI models over corporate data fears.
Three of the biggest enterprise technology names in the US are now setting terms for the two biggest AI labs. Nvidia, Palantir and Booz Allen Hamilton have either restricted how their staff use advanced models from Anthropic and OpenAI or threatened to drop them altogether, unless the labs can promise that proprietary corporate data will not be stored, read or absorbed into future training runs. The Information reported the pushback on Monday, and the specifics suggest this has moved well past a routine procurement complaint.
Palantir has asked Anthropic for an irrevocable zero data retention guarantee before it will make Claude available inside its own software. Nvidia has confined Anthropic’s models to internal work it considers low risk and routes anything proprietary to its own Nemotron models instead. Booz Allen Hamilton has barred employees from running Anthropic’s commercial model on cybersecurity projects that touch client data. Rather than wait, the two have shipped a product. Last week they unveiled a platform that pairs Palantir’s software with Nvidia’s open Nemotron models, letting organisations customise on their own operational data without it leaving their control. Nvidia is the first customer, applying it to its own supply chain. The commercial logic is plain. Every enterprise that cannot get a retention guarantee is a prospect. Microsoft is working the same anxiety, pitching isolated cloud environments that send nothing to an outside model provider.
The trigger is a policy change Anthropic made in June alongside Claude Fable 5, its first widely available Mythos-class model. Prompts and outputs on covered models are held for 30 days, and anything flagged by the safety classifiers can sit for up to two years. The Verge reported in June that Claude Fable 5 was missing from the model picker used by Microsoft’s own staff on internal GitHub Copilot, even as the company shipped the same model to Copilot and Foundry customers, while its lawyers worked through the retention terms.
Anthropic’s reasoning is that a single request looks harmless in isolation, and that attacks such as best-of-N jailbreaking or state-backed espionage only become visible when hundreds of requests are analysed together. Microsoft was the first large customer to flinch. The hesitation has spread since. Northrop Grumman runs open-source models on air-gapped servers. Novo Nordisk still uses Claude but keeps proprietary data away from it. One large US utility abandoned a Fable test for its core power infrastructure after Anthropic declined to sign a non-revocable zero data retention clause.