OpenAI keeps Zero Data Retention while expanding safety checks
Reviewed by Helen Jones
OpenAI has previewed Private Safety Processing for eligible API customers using Zero Data Retention (ZDR). The company says the system can spot risk patterns across related interactions while keeping prompts and model responses unavailable to OpenAI personnel; a broader rollout and technical paper are planned for September.
For teams that chose ZDR because their API workloads handle sensitive material, a safety feature that needs cross-interaction signals would normally reopen the retention question. This preview makes the architecture worth watching because it aims to add broader automated checks without turning prompts into routinely reviewable provider data. That does not make the privacy decision automatic. The preview is still being tested, and the practical boundary will depend on eligibility, the chosen storage setup, what limited safety signals are returned, and the organisation's own incident process.
Analysis
If you run eligible ZDR workloads, ask your security and platform owners to compare the September technical details with your retention promise, encryption-key ownership, alert handling and escalation path before enabling any new safety-processing option.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "OpenAI keeps Zero Data Retention while expanding safety checks", Collab365 Spaces.