OpenAI adds an AI safety layer to detect misuse without retaining enterprise data

OpenAI is introducing Private Safety Processing, a new safety capability enabling enterprises to detect AI system misuse across multiple interactions. This system identifies risk patterns without requiring OpenAI to retain prompts or model responses, upholding its Zero Data Retention (ZDR) commitments. The tool ensures OpenAI personnel cannot access underlying content during monitoring. Currently, the capability is being tested with eligible API and enterprise customers to address existing limitations in risk monitoring while preserving strict data privacy standards.