Anthropic Unveils EFS for Private, Cross-Session Misuse Detection

Companies deploying generative AI increasingly demand zero-data-retention, or ZDR, arrangements to limit exposure of sensitive information and meet compliance obligations. That privacy model, however, can make persistent abuse harder to detect because providers cannot connect suspicious activity across separate sessions. Anthropic’s Enterprise Frontier Safeguards, or EFS, is designed to bridge that gap, combining customer control of monitoring data with safeguards intended for more capable frontier models.

Anthropic said EFS will store monitoring data inside cloud infrastructure controlled by the customer, while the AI company supplies the analysis needed to identify misuse across sessions. The architecture is intended to preserve ZDR protections without forcing enterprises to give up longer-term threat detection. Applications are expected to open in phases beginning in fall 2026. Anthropic has not disclosed pricing, the number of initial participants or a precise general-availability date.