What happened
OpenAI announced (Aug 19, 2026) it is previewing Private Safety Processing with early customers — a system that detects misuse patterns across related interactions (e.g., agent drift, coordinated abuse) while keeping content encrypted with customer-controlled keys and inaccessible to OpenAI staff, designed to remain compatible with Zero Data Retention API commitments.
Why it matters
Resolves a growing tension for regulated/security-sensitive enterprises between frontier-model safety monitoring (which typically requires content retention) and strict no-retention data-privacy requirements.
Applicability
ZDR-eligible enterprise API customers with agentic/long-horizon workloads should evaluate the preview as it becomes available.