OpenAI preserves zero retention for frontier models

OpenAI is previewing Private Safety Processing so eligible API customers can keep zero data retention while automated systems look for risky patterns across related interactions, with 2 storage paths: customer-controlled infrastructure today and a planned OpenAI-hosted option encrypted with customer-held keys. The OpenAI announcement says OpenAI personnel cannot access the underlying customer content and that rollout plus a technical white paper are planned for September. Teams comparing frontier monitoring’s 20% compute overhead should file the distinction: privacy-preserving safety telemetry is becoming a deployment requirement, not a reason to surrender data custody.