On August 19, 2026, OpenAI announced Private Safety Processing, a system designed to monitor for misuse of its models while retaining zero customer data for paying API users. The system uses automated agents to detect coordinated or repeated abuse across conversations without human review unless flagged. OpenAI positioned the move as a stronger privacy stance than rival Anthropic, which retains user data for 30 days on certain models for safety purposes.
