Palantir, Nvidia, Booz Allen restrict Anthropic and OpenAI models

Palantir Technologies, Nvidia, and Booz Allen Hamilton are restricting or threatening to drop advanced AI models from Anthropic and OpenAI unless the labs provide stronger data protections, according to Reuters. The Information first reported the story, citing people familiar with the matter.

The core concern is that model providers could retain or extract value from customers' sensitive corporate information. Palantir has pressed Anthropic to provide guarantees of zero data retention before making its models available through Palantir's software, according to Reuters. Nvidia restricts Anthropic's models to less sensitive internal tasks, preferring its own Nemotron models for proprietary work; Booz Allen, meanwhile, has forbidden staff from running Anthropic's commercial model on cybersecurity projects that touch proprietary data.

The pushback intensified after Anthropic introduced a 30-day data retention policy in June tied to the rollout of its Fable 5 model, saying it needed usage logs to detect sophisticated attacks that unfold across multiple sessions. Anthropic said at the time that it would not use the retained data to train its models, but the policy drew objections from enterprise customers, particularly those in regulated industries. OpenAI has faced related scrutiny over claims it may have trained models on user data, according to Reuters.