OpenAI Launches Private Safety Processing for Enterprise AI

OpenAI announced on Wednesday a new privacy and safety framework called Private Safety Processing, designed to maintain its zero data retention promises for enterprise customers while strengthening its ability to detect misuse across multiple interactions. The system, which is in preview with early customers and expected to begin a broader rollout in September, aims to address a key tension in advanced AI deployment: safeguarding against complex threats without compromising user privacy.

Under the zero data retention model, OpenAI does not keep customer prompts or model responses after a request is processed, and enterprise data is not used to train models without explicit opt-in. The company stated that current safety systems typically evaluate each interaction independently. However, the most serious risks, such as coordinated hacking attempts or an AI agent acting beyond its intended authority, often only become apparent when examining patterns across a series of exchanges. Private Safety Processing is built to analyze these related interactions while keeping the underlying content inaccessible to OpenAI personnel.

For deployments using this framework, customer content will...