← registry

Claude Opus 5 Helped Researchers Take Over OpenAI Staff Accounts via Chained Flaws

Security researchers used Claude Opus 5 to chain multiple vulnerabilities in OpenAI's systems, gaining unauthorized access to employee accounts and internal code repositories through flaws in the help forum software and login system.

Categorymisuse
Severityhigh
AI systemchatbot
Sectorstechnology
Harm typessecurityprivacy
Lifecycle stageoperation
Actorexternal_attacker
Published2026-09-19 18:36:53

Summary is Secursion's own; full text lives at the source. Attribution preserved.