Nvidia and Palantir curb Anthropic Fable use over data-retention fears

Reuters, citing The Information and people familiar with the matter, reports that large technology companies including Palantir Technologies, Nvidia, and Booz Allen Hamilton could restrict or stop using advanced AI models unless Anthropic and OpenAI guarantee they will not misuse customer intellectual property — the pushback tied to Anthropic’s June policy for its Fable model that retained usage logs for 30 days to help detect “complex and novel attacks.”

While CEOs debate pacing in public, big buyers are already throttling the hottest models over who keeps the logs. Fable’s 30-day retention fight is about zero-data-retention — a promise the lab will not keep your prompts — when the model sees company code and secrets.

Reuters’ Prathik Jayaprakash, writing from Bengaluru on 14 Sep 2026 at 3:11 PM UTC, reported that large technology companies — including Palantir Technologies, Nvidia, and Booz Allen Hamilton — could restrict or cease using advanced AI models unless Anthropic and OpenAI guarantee they will not misuse those customers’ intellectual property. Intellectual property, or IP, here means a company’s own code, documents, and secrets. Reuters cites The Information and people familiar with the matter. That Reuters pickup is the filing event. This is unnamed-source reporting. It is not a company announcement, not a signed industry pact, and not a claim that the named buyers have already shut every model off for good.

Fable, as used here, is Anthropic’s newer, more capable Claude-family model line — the one the June policy change was written around. Reuters says Anthropic faced customer pushback after that June policy for Fable gave the company rights to retain usage logs for 30 days to help detect “complex and novel attacks.” Usage logs are the record of what a customer typed in and what the model wrote back.

Zero-data-retention, or ZDR, is a contract promise that the lab will not keep those prompts and answers after it replies. Reuters says Palantir pressed Anthropic to provide irrevocable ZDR guarantees before making Anthropic models available through Palantir’s own software, citing a person familiar with the discussions. This is not a claim that Palantir already shipped those models, or that a ZDR contract was signed.

Nvidia, the same report says, limits Anthropic to less sensitive tasks and relies on its own Nemotron models for internal work. Nemotron is Nvidia’s in-house model family. This is a reported throttle — not a claim that Nvidia banned Anthropic everywhere, and not a product review from this desk.

Booz Allen Hamilton, the report says, has barred employees from using Anthropic’s commercial model for proprietary cybersecurity work. Proprietary here means the firm’s own, in-house security work, not a public demo. This is not a claim that Booz Allen banned every Anthropic tool for every job.

Microsoft, Reuters says, is capitalizing on the same fears by pitching isolated cloud environments and its own AI offerings to worried clients, according to the report that cited people familiar with the matter. Isolated cloud means a private computing room that is walled off from the public internet and from other customers. This is sales talk in a tip, not a named Microsoft contract this desk can cite.

Both labs, the report adds, say they do not train on customer data by default unless companies opt in, though they collect anonymized metadata to improve products. Train on customer data means using what a client typed to teach the next model. Metadata means leftover facts about the traffic — counts, times, sizes — not the full secret text.

Same Reuters dispatch: OpenAI has faced similar scrutiny amid claims it may have trained models on user data to help solve a math problem.

Palantir, Nvidia, Booz Allen, Anthropic, and OpenAI did not immediately respond to Reuters requests for comment. Absence of a reply is not confirmation.