OpenAI Launches Private Safety Processing Preview With Zero Data Retention
Private Safety Processing extends safety monitoring from individual interactions to full patterns without breaking the Zero Data Retention promise, addressing the tension between abuse detection and data privacy for enterprise API customers.
Reporting from 1 source: GIGAZINE.
OpenAI released a preview of Private Safety Processing on August 19, 2026. The system lets OpenAI detect misuse patterns across related interactions without staff accessing customer content, while keeping the Zero Data Retention policy intact. Customer content stays on customer-managed infrastructure or encrypted storage with customer-held keys.
OpenAI's existing Zero Data Retention policy evaluates each interaction separately, which limits how well automated systems can spot coordinated misuse. Private Safety Processing builds on the same automatic protections but expands them across related interactions, letting OpenAI identify patterns without staff ever seeing the underlying prompts or responses.
The system works whether customer content sits on customer-managed infrastructure or on OpenAI-provided storage. In the latter case, content is encrypted with customer-managed keys, and OpenAI staff hold no copies of those keys. When a risk is flagged, OpenAI receives only a limited signal describing the type of activity, which feeds decisions about whether legal action is warranted.
Customers can investigate alerts using information available in their own systems, and may choose to share relevant details with OpenAI for appeals or cooperation on verified misuse. The preview is currently being tested with initial customers.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.