Private Safety Processing
OpenAI's system for spotting misuse across several conversations while still storing none of the content.
Previewed in August 2026, Private Safety Processing is OpenAI’s attempt to have it both ways: keep Zero Data Retention for business customers while still noticing patterns of abuse that only show up across several related interactions. Instead of the prompts and answers, the system passes along a narrow signal describing what type of activity it saw and how severe it looked. OpenAI staff never see the underlying text.
It exists because AI is moving from single questions to agents that run long chains of steps on their own, and a safety check that only ever sees one message cannot spot a pattern. The feature is aimed at enterprise and API customers rather than consumer ChatGPT plans, with a rollout and a technical white paper both expected in September 2026.