How Indian-origin techies tricked Claude into writing an exploit to hack OpenAI?

An ethical hacking team led by Indian-origin researchers successfully bypassed Claude Opus 5's safety guardrails during an authorised OpenAI bug bounty assessment. When the AI initially refused to write exploit code for a real target, the team disguised their test server as a capture-the-flag training challenge. Believing it was an ethical training practice, Claude generated the exploit.