Offering Zero Data Retention for frontier models

OpenAI is previewing a system called Private Safety Processing that allows it to monitor for misuse across multiple customer interactions while keeping the underlying content hidden from OpenAI personnel. The system works by analyzing patterns in customer data without exposing the actual prompts or responses, using automated signals to flag potential risks instead. This matters because as AI systems handle longer and more complex tasks, safety risks often only become visible when viewing multiple interactions together, but many organizations need to keep their sensitive data private due to regulatory obligations. The new system is designed to let OpenAI continue offering Zero Data Retention, which means customer prompts and responses are not retained after processing, while still maintaining the ability to detect harmful patterns that single interactions alone would miss.