OpenAI said it will tighten safety processes for paying customers and has begun testing "private safety processing", a system designed to detect risk patterns across multiple interactions rather than individual prompts and replies. The move targets risks that emerge as AI handles longer-horizon, more autonomous tasks and is intended to guard against malicious users and attacks by misaligned AI agents while remaining compatible with zero-data-retention—customer content will not be accessible to O

OpenAI said it will tighten safety processes for paying customers and has begun testing "private safety processing", a system designed to detect risk patterns across multiple interactions rather than individual prompts and replies. The move targets risks that emerge as AI handles longer-horizon, more autonomous tasks and is intended to guard against malicious users and attacks by misaligned AI agents while remaining compatible with zero-data-retention—customer content will not be accessible to OpenAI. The feature is expected to roll out in September, accompanied by a technical white paper; early tests are underway with Microsoft and Databricks.