NOISESaved copy
Coverage history

Assembled by NOISE.

Saved report copy

Recorded

Researchers used Claude to hack OpenAI employee accounts

A three person team at Hacktron AI chained two vulnerabilities starting with a HEIF image flaw in libheif on OpenAI's Discourse forum to take over employee ChatGPT and Codex accounts and demonstrate access to an internal repository, earning a $6,500 bounty after OpenAI fixed the issue. TC frames the incident as evidence that off the shelf AI lowers the expertise needed for exploits, noting the team succeeded within hours of switching from Claude Opus 4.8 to Opus 5, while Register emphasizes the less than 72 hour timeline and OpenAI's clarification that the award covered the OpenAI side finding, not the Discourse host.