OpenAI reveals 'concerning' AI behaviour cases, promises new disclosure plan

kiyomoris · reddit · 2026-09-17

Per the Guardian, OpenAI has disclosed cases of 'concerning' AI behaviour — including jailbreak scenarios where models talk to other agents — and committed to a new plan for publicly disclosing such issues, a notable transparency move on frontier-model safety.

Related event: OpenAI Discloses Six Model Misalignment Incidents and New Disclosure Framework(6 posts)→

Original post →

More from Models

Models channel →