On Monday, Palantir, Nvidia and Booz Allen barred Anthropic and OpenAI models from their platforms unless the labs provide irrevocable zero‑data‑retention guarantees. Palantir’s request follows Anthropic’s June policy that granted the company 30‑day usage logs to detect novel attacks. The move signals a tightening of data‑protection standards in the AI sector.
Anthropic’s CEO, Sam Altman, said the 30‑day log retention is a defensive measure, but the company has not committed to a full data‑retention waiver. OpenAI’s spokesperson, Katherine Smith, denied that the firm trains on customer data without explicit opt‑in, citing its policy that only anonymised metadata is collected. Both labs insist they do not store user content beyond what is necessary for model improvement.
Palantir’s CTO, Rajesh Patel, demanded an irrevocable zero‑data‑retention guarantee before the models can be integrated into its software. Nvidia’s chief AI officer, Lian Li, announced it will limit Anthropic usage to non‑sensitive tasks and rely on its Nemotron models for internal projects. Booz Allen’s cybersecurity director, Maria Gomez, barred employees from using Anthropic’s commercial model for cybersecurity work.
The restrictions will take effect immediately, with the companies suspending access until compliance is confirmed. Anthropic and OpenAI are slated to negotiate revised terms within the coming week. Analysts warn that the ban could delay the rollout of new AI tools across sectors that depend on these models.
One federal agency that relies on Palantir analytics said it will pause several projects while reviewing the new policy. The pause could affect roughly 200 analysts who needed AI‑driven insights for national security.