Palantir, Nvidia and Booz Allen restrict Anthropic's Fable over data-retention terms
The Information reported on Monday, September 14, 2026 that Palantir, Nvidia and Booz Allen Hamilton are restricting Anthropic's Fable model over a 30-day data-retention window.

The fight is over 30 days. Anthropic keeps prompts and outputs from Fable, the generally available frontier model it launched in June for coding and knowledge work, for 30 days so it can spot misuse across sessions. Palantir Technologies, which sells data software to governments and companies, has offered Claude inside its AI platform on Amazon Web Services for U.S. intelligence and defense work since November 2024. Nvidia, the AI chip designer, committed in November 2025 to invest up to $10 billion in the lab. The Information reported Monday that Palantir, Nvidia, and Booz Allen Hamilton, the U.S. government contractor, are restricting Fable for sensitive work over those retention terms.
Reuters, citing The Information, said the three companies could restrict or cease using advanced AI models unless Anthropic and OpenAI guarantee they will not misuse the companies' intellectual property, and that OpenAI has faced similar scrutiny over claims it may have trained models on user data. Palantir has pressed Anthropic for an irrevocable zero-data-retention guarantee before putting the models through Palantir's software, Nvidia limits Anthropic models to less sensitive tasks and uses its own Nemotron models internally, and Booz Allen has barred employees from using Anthropic's commercial model on proprietary cybersecurity work, Reuters said. Palantir, Nvidia, Booz Allen, Anthropic, and OpenAI did not immediately respond to Reuters.
Anthropic's published terms say the 30-day data is not used for training without permission. A zero-data-retention arrangement, under which Anthropic does not store prompts or answers after it replies, exists for other products, but Fable is off that arrangement unless Anthropic expressly authorizes it. Some customers want that guarantee to be irrevocable before they put proprietary information into the model, The Information said. On September 1 Anthropic said it would replace the 30-day hold with Enterprise Frontier Safeguards, which keep monitoring data in the customer's own cloud, rolling out in phases later this fall, with eligible customers getting temporary zero-data-retention treatment on Fable 5 and Fable 5.1 until then.
You cannot buy Anthropic. Harbor's Anthropic AI Lab Ecosystem ETF ANTW, which holds public companies the adviser links to the lab, had $9.01 million in net assets as of September 10, Harbor reported, and tells holders not to expect a direct stake in Anthropic. The next dated step is whether that fall safeguards system actually lands for the customers who will not use Fable without it.
Frequently asked
Why are these companies restricting Fable?
Anthropic retains prompts and outputs for 30 days to spot misuse across sessions, and the customers do not want proprietary information sitting in that window.
Is that retained data used to train Anthropic's models?
Anthropic's published terms say the 30-day data is not used for training without permission.
Can customers get zero data retention on Fable?
Zero retention exists for other Anthropic products, but Fable is off it unless Anthropic expressly authorizes it.
Is OpenAI facing the same pressure?
Reuters, citing The Information, said OpenAI has faced similar scrutiny over claims it may have trained models on user data.
Can investors buy Anthropic directly?
No: the Harbor ETF linked to the lab holds public companies the adviser associates with Anthropic and tells holders not to expect a direct stake.