OpenAI pitches Private Intelligence for sensitive work

This is a meaningful enterprise trust layer, but its privacy promise currently spans two different guarantees: customer-controlled safety retention is available, while private inference remains unfinished.

• –Private Safety Processing stores selected records in customer-controlled storage with a 30-day TTL and restricts decryption to a hardware-attested runtime. [OpenAI API docs](https://developers.openai.com/api/docs/guides/private-safety-processing)
• –Enterprise Key Management gives customers control over decryption authorization, adding a practical layer of governance.
• –Private Inference is the more important differentiator for highly sensitive workloads, but its supported models, endpoints, pricing, and verification details remain unclear.
• –Developers must check endpoint eligibility carefully: several agent-related APIs remain outside Zero Data Retention coverage. [Data controls](https://developers.openai.com/api/docs/guides/your-data)