TODAY'S SIGNAL
OpenAI is previewing a system for eligible zero-data-retention API deployments that it says can look for risky patterns across related interactions while the underlying content stays in customer-controlled infrastructure. OpenAI personnel would receive a limited risk signal, not the prompts or responses themselves.
Why it matters: organizations handling sensitive information have often faced a blunt tradeoff—give the AI provider more visibility for safety monitoring or keep tighter control of their data. This preview tries to separate those two functions.
The honest tension: this is a vendor-described design in early customer testing, not an independently validated control. OpenAI says a technical paper and rollout details are planned for September, and its legal exception for apparent child sexual abuse material remains.
The honest limit
This is an early vendor preview, not an independently validated control.
The technical paper, the exact eligibility and storage terms, independent security review, false-positive handling, and whether customers can audit enforcement without disclosing the content they were trying to protect.
Public information only. Not professional advice.
QUICK FEEDBACK
Was this relevant to you?