White hat hackers breach OpenAI using Anthropic's Claude

Researchers from Hacktron AI successfully accessed OpenAI's internal repository by exploiting vulnerabilities in the community.openai.com forum and OpenAI's single sign-on system. The team utilized Anthropic's Claude models to identify security gaps in the libheif library and develop a functional remote code execution exploit. The breach occurred on July 25, and the findings were reported to OpenAI and Discourse the same day, with full details disclosed on September 13.