Security researchers at Hacktron AI used Anthropic's Claude Opus 5 to exploit vulnerabilities in OpenAI's systems, gaining access to employee ChatGPT accounts through a flaw in Discourse's image upload handling. OpenAI awarded the team a $6,500 bug bounty after the vulnerabilities were reported and subsequently patched.