Nvidia, Palantir restrict Anthropic models over data-retention fears, says The Information
mark_k · x · 2026-09-15
Per The Information, major customers are restricting Anthropic models over data-retention concerns:
- Nvidia limits Anthropic models to less sensitive tasks
- Booz Allen barred its commercial model from proprietary cybersecurity work
- Palantir demands an irrevocable zero-data-retention guarantee
Anthropic says it doesn't train on corporate customer data unless customers opt in. But that doesn't address the core worry: whether sensitive data is retained at all. For companies protecting valuable IP, a promise not to train on data is different from a guarantee not to keep it.
More from Companies & People
- Silicon Valley hearsay: security researchers will soon out-earn AI researchers — eliza_luth · 2026-10-31
- Dev slams Copilot but praises GitHub execs after candid Universe chats — DanWahlin · 2026-10-31
- Sequoia's Maguire pushes back on Ed Zitron's OpenAI jab: "People have no idea what's coming" — shaunmmaguire · 2026-10-31
- Stratechery insight: Palantir's FDE is a product strategy, not just a role — annetgriffin · 2026-10-31
- Google's Web AI Summit 2026 draws 1K+ RSVPs with 24 speakers, 25 edge-AI demos — jason_mayes · 2026-10-31
- Proton CEO cancels forced AI "re-interviews" after exposé; users call for resignation — 404 Media · 2026-10-31