Anthropic’s Claude assisted researchers in breaching OpenAI systems

Security researchers from Hacktron utilized Anthropic’s Claude to identify vulnerabilities and breach OpenAI accounts within 72 hours. By chaining an image-processing flaw with an identity infrastructure issue, the team gained access to internal GitHub repositories. The researchers ceased testing after creating a harmless pull request and reported the findings to OpenAI, which patched the flaw and awarded a $6,500 bounty. The attack was notably accelerated by the release of Claude Opus 5, which successfully solved an exploitation hurdle that previous models could not overcome.