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

Researchers exploited two distinct vulnerabilities in sequence — a flaw in the libheif library used by OpenAI's Discourse-based help forum, combined with a weakness in OpenAI's own authentication system — to gain unauthorized access to employee accounts and an internal code repository. This attack demonstrates the danger of vulnerability chaining, where individually moderate flaws become critical when combined. Third-party software components (like libheif embedded in Discourse) expand an organization's attack surface in ways that are easy to overlook. The incident underscores that public-facing support and community platforms are not low-risk assets — they can serve as pivot points into core internal systems.