OpenAI's proposal is a material privacy-and-safety control, but its operational effectiveness remains unvalidated because the feature is still in early customer testing.
Limits of the evidence: The feature is described by OpenAI as a preview and is being tested with early customers; the source provides no independent evaluation of detection accuracy, privacy guarantees, or operational performance.
What to watch
Taken from the first report; the whole story has not been re-assessed yet.
Offering Zero Data Retention for frontier models
OpenAI NewsAug 19, 2026First report
No revisions yet.
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.