Indian researchers use Claude to hack OpenAI in security test

Three Indian-origin researchers used Anthropic's Claude to hack into OpenAI's systems during an authorized security test in July, gaining access to employee accounts and a private GitHub repository. The team spent under $3,000 on AI tokens and received a $6,500 bug bounty. The researchers—Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini—work at cybersecurity startup Hacktron AI. They exploited a flaw in OpenAI's public community forum involving image files, then used Claude to write and debug the exploit, completing the test in under 72 hours. The trio has extensive bug-bounty experience, with backgrounds at companies like Cure53, Zomato, and HackerOne. Pedhapati is Hacktron's CTO, Jaiswal led the research, and Maini has received multiple security community awards. OpenAI has since fixed the vulnerabilities.