OpenAI previews privacy-preserving safety monitoring for Zero Data Retention customers

OpenAI is previewing Private Safety Processing, a system intended to give eligible API customers stronger safety monitoring while preserving Zero Data Retention (ZDR) commitments.

Under ZDR, OpenAI says prompts and model outputs are not retained after processing, are unavailable to its personnel for review, and are not used for training unless a customer opts in. The company says the new system will extend automated safety checks beyond individual requests, looking for patterns across related interactions that could indicate misuse, coordinated probing, or an AI agent acting beyond its authority.

What this means for builders

The approach is aimed at organizations using frontier models for sensitive workloads, including healthcare, finance, research, and proprietary business operations. Customer content remains in infrastructure controlled by the customer for ZDR deployments. OpenAI is also developing an alternative using its own storage, with customer-controlled encryption keys that OpenAI staff cannot access.

When automated systems detect a concern, OpenAI says it will receive only a limited signal describing the activity and its severity—not the underlying prompts or responses. Customers can review alerts and enforcement decisions through their own systems, and may voluntarily share relevant material when appealing a decision or helping investigate confirmed abuse.

The system is currently being tested with early customers. OpenAI plans to begin rollout and publish a technical white paper in September, giving developers more detail before they assess the feature for production use.

One important exception remains: images flagged as potential child sexual abuse material may still be retained for required human review and reporting, including in ZDR deployments.

Source: OpenAI News