Offering Zero Data Retention for frontier models

THE NEWS

What happened

OpenAI is previewing Private Safety Processing, a new capability designed to maintain its Zero Data Retention (ZDR) commitment while enabling automated safety systems to detect patterns across multiple interactions. This addresses emerging risks in complex AI tasks where single-interaction monitoring is insufficient, allowing OpenAI to flag potential misuse without personnel accessing customer content.

CONTEXT

Why it matters

OpenAI launches Private Safety Processing preview: a new safety layer that detects patterns across multiple AI interactions without accessing customer data. This preserves Zero Data Retention commitments while addressing risks in complex agentic tasks. Content remains encrypted or on customer infrastructure, with OpenAI receiving only anonymized risk signals. A technical white paper is expected in September.

AT A GLANCE

Key facts

The main verified points:

• OpenAI introduces Private Safety Processing to identify harmful patterns across related interactions without human access to customer data.
• The feature extends existing Zero Data Retention (ZDR) protections, ensuring prompts and responses are not retained or reviewed by OpenAI staff after processing.
• Customer content remains on infrastructure controlled by the customer or is stored on OpenAI infrastructure encrypted with keys held exclusively by the customer.
• When risks are detected, OpenAI receives only a narrowly defined signal indicating the activity type, never the underlying content.
• The system is currently in preview with early customers, with a technical white paper and broader rollout planned for September.
• CSAM detection remains an exception where images may be retained for legal reporting obligations, consistent with existing laws.