Overview
- This week OpenAI announced Zero Data Retention (ZDR) and previewed Private Safety Processing, and it says both are being tested with early enterprise customers ahead of a broader rollout and a September technical white paper.
- ZDR promises that eligible deployments will not retain customer prompts or model responses after a request and will not use enterprise data to train models unless a customer opts in.
- Customers can store content on their own infrastructure or on OpenAI infrastructure encrypted with customer-controlled keys so OpenAI staff do not hold the keys, and some chat endpoints will enforce store=false to stop routine logging.
- OpenAI explicitly allows an exception for images flagged as possible child sexual abuse material to be retained for manual review and reporting, and it warns some operational metadata or application state may still be kept for service reasons.
- Key practical details remain unpublished — which endpoints and regions qualify, the exact scope and semantics of the safety signals, exception handling, and pricing — so regulated organizations must verify contracts and test configurations before deployment.