OpenAI Expands Zero Data Retention with Private Safety Processing

OpenAI is expanding its Zero Data Retention service to eligible frontier model API customers. This means once a request is processed, OpenAI doesn't keep the prompts or responses, and employees can't review customer content. Your data won't be used for training unless you opt in.

As models handle longer and more complex tasks, some safety risks might only show up after multiple interactions. So OpenAI is introducing Private Safety Processing as a preview. This system automatically looks for risk patterns across linked interactions without exposing the actual content to OpenAI staff. It works whether customers run their own infrastructure or use OpenAI's storage, where content is encrypted with customer-held keys.

When risks are detected, OpenAI only gets limited safety signals about the activity type, not the content itself. Customers can investigate alerts using their own systems and can provide info to OpenAI if they want to appeal or verify legitimate use. The feature is being tested with early customers, with a rollout planned for September along with a technical whitepaper.