From：Internet Info Agency 2026-08-20 07:34:00
OpenAI announced today an expansion of its "Zero Data Retention" service to customers using its frontier model APIs. This service ensures that no prompts or model responses are retained after request processing is complete, that employees cannot review customer content, and that enterprise data will not be used to train models unless the customer explicitly opts in. Additionally, OpenAI has launched a preview of Private Safety Processing, designed to identify potential safety risks across multi-turn interactions. This mechanism uses automated systems to analyze patterns of risk across related interactions without accessing the underlying content. The system works both for content managed within customer-controlled infrastructure and for content stored using OpenAI’s storage services; in the latter case, customer content is encrypted with customer-controlled keys, and OpenAI employees cannot access the underlying prompts or responses. When the system detects a risk, OpenAI receives a limited-scope safety signal indicating the type of activity involved, which it uses to determine whether enforcement action is necessary. Even when content is flagged, employees cannot view the customer’s actual content. Customers can investigate alerts and enforcement decisions through their own systems and may choose to provide information to OpenAI to appeal or assist in verifying confirmed abuse. OpenAI plans to begin rolling out this capability in September and will release a technical whitepaper detailing the approach.
