Offering Zero Data Retention for frontier models

OpenAI is expanding its Zero Data Retention offering for API customers, guaranteeing that prompts and responses are not stored or accessible to OpenAI personnel after processing, with enterprise data excluded from model training unless customers opt in. To address a gap in safety monitoring, OpenAI is previewing Private Safety Processing, a system designed to detect harmful patterns across multiple interactions without exposing underlying content to OpenAI staff. The system works by analyzing encrypted customer content and returning only narrow, predefined safety signals when risks are identified, keeping the actual data under customer control. OpenAI-provided storage options will use encryption keys held exclusively by the customer, meaning OpenAI personnel cannot access flagged content directly. Private Safety Processing is currently in early testing and is scheduled to launch with a technical white paper in September 2025.