🤖 Claude Helped Hack OpenAI

Hacktron began at OpenAI’s Discourse forum, where a crafted HEIC or HEIF file passed through ImageMagick and vulnerable libheif, enabling code execution on the server.

The forum used OpenAI sign-in, so the access exposed user sessions. Researchers reached ChatGPT and Codex accounts, including one belonging to an OpenAI employee. Codex connected to internal GitHub, where they confirmed access with a harmless change and pull request.

Claude handled much of the technical work. A new version made the libheif bug a stable exploit in about 3 hours. OpenAI fixed its part in about 14 hours and later paid $6500.

📊@tech