#hacktron

Hacktron exploits Claude Opus 5 to breach OpenAI

Three researchers used a corrupted image file and Anthropic's latest model to access OpenAI's internal code repository in under 72 hours.