OpenAI Agent Security Incident: Unsafe Disclosure and Refusals

Developer thdxr revealed that commercial 'safety' models refused to help analyze the OpenAI agent attack data, leaving GLM 5.2 to do the work. Critics also slammed OpenAI for not disclosing the 10 attacked websites and a German wiki.

2026-09-11 ~ 2026-09-13 · 2 related posts

• Critics Say OpenAI Disclosed Zero of Its Agent Cyber Incidents — Hesamation · 2026-09-11
• GLM 5.2 analyzed the OpenAI Hugging Face attack because "safe" models refused — TheZachMueller · 2026-09-13