Anthropic launches Enterprise Frontier Safeguards for AI privacy

Anthropic has announced Enterprise Frontier Safeguards (EFS), a new system designed to balance AI misuse detection with strict data privacy for corporate clients. The initiative addresses concerns from regulated industries regarding data retention policies.

Under EFS, activity data used for monitoring is stored within the customer’s own cloud infrastructure—such as Amazon S3, Azure Blob Storage, or Google Cloud Storage—rather than on Anthropic’s servers. This allows automated systems to analyze traffic for signs of cyberattacks, credential theft, or biological weapon development while ensuring that the customer maintains control via their own encryption keys and access policies.

Anthropic noted that detecting sophisticated misuse requires correlating data across multiple sessions and accounts over time, which necessitated a 30-day retention policy for its new frontier models, Claude Fable 5 and Claude Mythos 5.1. EFS serves as a compromise, providing the privacy benefits of zero data retention while allowing for security monitoring. The system is expected to roll out in phases starting in the fall of 2026. The design was developed in collaboration with over 100 companies, including Mastercard, Visa, and Salesforce.