Loading…
OpenAI reports a large reduction in protected-reasoning extraction success under its internal test, but the result needs independent validation.
Limits of the evidence: The attack-success figures are from an internal OpenAI evaluation; the source does not provide independent replication or full evaluation details.
What to watch
Assessment revised Nov 20, 2026
Combatting adversarial distillation
OpenAI NewsNov 20, 2026First report
First assessment, from OpenAI News
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.