Cybersecurity researchers successfully breached internal OpenAI infrastructure by utilizing Anthropic's Claude chatbot to chain multiple vulnerabilities together during an authorized security audit.
The security team from the startup Hacktron AI uncovered a flaw in the image processing software running OpenAI's community discussion forum. By leveraging Claude to write and refine custom exploit code, the researchers obtained authentication tokens belonging to internal OpenAI employees. The compromised credentials granted the team unauthorized reach into OpenAI's ChatGPT accounts and internal GitHub source code repository.
Hacktron reported the vulnerabilities directly to OpenAI through its official bug bounty program, earning a six thousand five hundred dollar reward after the flaws were resolved. The incident underscores growing concerns that advanced artificial intelligence agents can drastically reduce the technical difficulty and time required to execute complex cyberattacks.