OpenAI reaffirms its Zero Data Retention (ZDR) commitment for eligible API customers, meaning prompts and responses are not retained or used for training without explicit opt-in. It previews a new feature called Private Safety Processing, designed to detect misuse patterns across multiple related interactions without exposing underlying content to OpenAI staff. Customer data stays on customer-controlled infrastructure or is encrypted with customer-held keys when stored by OpenAI; only narrow risk signals are surfaced. The feature is in early customer testing, with rollout and a technical white paper planned for September, alongside a supporting quote from Glean.
Table of contents
Why safety systems need to evolveHow Private Safety Processing worksPrivacy and safety built with and for our customersQuestions this post answers
What is OpenAI's Private Safety Processing feature?
Private Safety Processing is a preview feature from OpenAI designed to detect misuse patterns across multiple related interactions without giving OpenAI personnel access to the underlying prompts or responses. It extends existing Zero Data Retention (ZDR) compatible safety systems, which evaluate interactions individually, by analyzing patterns across sessions using automated systems that only return narrow risk signals. It is being tested with early customers, with rollout and a technical white paper planned for September. Enterprise teams weighing AI vendor data handling can track rollouts like this one on daily.dev.
Does OpenAI's Zero Data Retention policy stop it from detecting misuse across multiple interactions?
Existing Zero Data Retention compatible safety systems only evaluate each interaction individually, which limits detection of risks that only become visible when multiple interactions are viewed together, such as bad actors probing safeguards over time or an agent continuing to act after being told to stop. OpenAI's new Private Safety Processing feature is built to close that gap while preserving ZDR guarantees. daily.dev helps developers building on ZDR-style APIs keep up with how safety monitoring evolves.
How does OpenAI protect customer content when using OpenAI-provided storage instead of Zero Data Retention?
When customers use OpenAI-provided storage rather than keeping content on their own infrastructure, the content is encrypted using keys controlled by the customer, and OpenAI personnel do not hold copies of those keys, so they cannot access the underlying content. Automated systems can still flag potential misuse and return a narrowly defined signal about the type of activity without exposing the actual prompts or responses. Teams choosing between self-hosted and vendor-hosted AI storage can follow these tradeoffs on daily.dev.