OpenAI has announced that it will continue offering Zero Data Retention (ZDR) for eligible API customers, now explicitly covering its frontier models. The policy ensures that OpenAI does not store prompts or outputs after processing, a key requirement for enterprises in regulated industries. The announcement also previews a new feature called Private Safety Processing, which is designed to apply advanced safety filters without requiring the data to be retained or inspected by humans.
Private Safety Processing appears to be a technical workaround that lets OpenAI run safety evaluations on model behavior while keeping customer content ephemeral. This addresses a long-standing tension: safety monitoring often implies logging, but ZDR customers want neither storage nor review. OpenAI's approach suggests a shift toward algorithmic, in-memory checks that produce only aggregate safety signals.
The source does not specify which models qualify as "frontier" or when the private safety processing will be generally available. It also does not detail how the safety checks differ from standard moderation. Still, the announcement signals that OpenAI is treating data privacy as a competitive feature for its most advanced offerings, not just a compliance checkbox.