Offering Zero Data Retention for frontier models

OpenAI is introducing Private Safety Processing, a system designed to identify patterns of harmful behavior across multiple AI interactions while keeping customer data private under Zero Data Retention policies. Existing safety systems evaluate each interaction individually, but as AI models take on longer and more complex tasks, serious risks may only become visible when multiple interactions are viewed together, such as when bad actors repeatedly probe safeguards or coordinate across accounts. Private Safety Processing uses automated systems to detect these patterns without giving OpenAI personnel access to the underlying customer content, whether that content is stored on customer-controlled infrastructure or on OpenAI infrastructure encrypted with customer-controlled keys. This approach addresses a conflict some organizations face: other frontier model providers have required data retention for safety monitoring, which conflicts with customer security obligations, while OpenAI's solution aims to advance safety capabilities without compromising privacy and control.