OpenAI has introduced Private Safety Processing (PSP), a system designed to monitor safety risks in multi-turn AI interactions without storing user data. The tool ensures zero data retention for API and enterprise clients, meaning information is neither saved, reviewed, nor used for model training.
The system transmits only limited safety signals to OpenAI, keeping the interaction content private. Customers can choose to keep data entirely within their own infrastructure or store it encrypted on OpenAI servers, with the client retaining control over the encryption keys.
Announced on August 19, 2026, the system began a phased rollout to API customers on September 22, 2026. Microsoft, Databricks, Glean, and Abridge participated as early development partners. A technical white paper detailing the architecture was released alongside the tool.
OpenAI positions PSP as an alternative to retention-based models, such as the 30-day data retention policy implemented by Anthropic in June 2026. The new feature is currently exclusive to enterprise and API users and is not available for consumer ChatGPT plans.
