Offering Zero Data Retention for frontier models
OpenAI introduces Private Safety Processing to strengthen safeguards across multiple user interactions while maintaining Zero Data Retention (ZDR) for eligible API customers, ensuring prompts and responses are not retained or reviewed by personnel.
OpenAI announced Private Safety Processing, a new capability designed to detect risks across related interactions without retaining or reviewing customer prompts or responses. This feature aligns with the company’s Zero Data Retention policy, which guarantees that eligible API customers’ data is not stored or accessible to OpenAI personnel after processing. Enterprise customers retain control over their data, with options for encrypted storage using customer-managed keys. The system identifies misuse patterns while maintaining strict privacy protections, addressing concerns about long-term safety risks in complex AI tasks.
Private Safety Processing extends existing ZDR-compatible safety systems by analyzing patterns across multiple interactions without exposing underlying content to OpenAI staff. It supports deployments where customer data resides either on customer-controlled infrastructure or encrypted storage provided by OpenAI. Automated systems generate limited safety signals when risks are detected, enabling enforcement actions without revealing sensitive information. Customers retain full visibility into alerts and can investigate or appeal decisions using their own systems, with optional sharing of relevant data for further review.
The announcement follows feedback from enterprise customers requiring predictable data protection as AI systems grow more capable. OpenAI emphasizes collaboration with industry partners to refine safeguards, noting that many organizations handle highly sensitive information subject to strict regulatory requirements. Private Safety Processing is currently in testing with early adopters and will be detailed in a technical white paper scheduled for September. The company plans to provide ongoing updates to customers to ensure transparency and facilitate integration.
OpenAI reiterates its commitment to reporting illegal content, such as apparent child sexual abuse material (CSAM), which will continue to be retained for manual review and reporting under ZDR deployments. The company frames Private Safety Processing as a response to the evolving needs of enterprise AI adoption, balancing advanced safety measures with stringent data privacy controls to maintain customer trust and regulatory compliance.