Overview
- Anthropic introduced a policy in June that lets its Fable model retain customer usage logs for 30 days and flagged items longer, a change that triggered enterprise alarm about intellectual‑property exposure.
- In response to the June policy, several major firms have limited or barred Anthropic and OpenAI models from sensitive work, with Palantir pressing for an irrevocable zero‑data‑retention guarantee and Nvidia routing proprietary tasks to its Nemotron models.
- Anthropic on Sept. 1 announced Enterprise Frontier Safeguards, a feature that stores flagged activity data in a customer’s own cloud under the customer’s encryption keys while automated safety checks run, with phased rollout planned for the fall.
- The dispute has already changed procurement and deployments, including at least one large U.S. utility canceling a Fable test and customers moving sensitive workloads to sovereign clouds, on‑premises setups, or locally hosted open models.
- The fight over logs and metadata is shifting the market: vendors such as Microsoft, Palantir and Nvidia are pitching isolated or customer‑controlled AI environments, forcing labs to supply clearer contracts and technical controls or risk losing regulated clients.