Indian cybersecurity experts successfully breached OpenAI using Anthropic’s Claude
Three Indian-origin cybersecurity experts successfully breached OpenAI systems by leveraging Anthropic’s Claude chatbot during an authorized security test. Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini from the startup Hacktron AI uncovered a significant software flaw in July. Their authorized intrusion granted them access to sensitive employee accounts and a restricted GitHub environment.
Exploiting system vulnerabilities with Claude
Bug bounty programs have become essential for tech giants seeking to fortify their networks against increasingly sophisticated cyber threats. Operating under this framework, the trio identified a vulnerability within an OpenAI public community forum. This specific flaw involved how the platform processed certain image files. It allowed outsiders to execute unauthorized code on the server.
To investigate and build a working exploit, the team turned to Claude. This rival AI assisted with writing and debugging the necessary code, significantly accelerating their workflow. The researchers discovered an additional weakness that exposed login tokens. These tokens provided a direct pathway into a private GitHub repository via an employee’s Codex account. While Claude generated the code, the researchers themselves connected the different vulnerabilities to orchestrate the final breach.
Executing this entire operation took the researchers less than 72 hours. They spent under $3,000, roughly Rs 2.5 lakh, on AI tokens to fuel Claude during the testing process. OpenAI swiftly patched the identified vulnerabilities and rewarded the Hacktron AI team with a $6,500 bug bounty, approximately Rs 5.5 lakh, for their discovery.
These three researchers bring extensive experience to the cybersecurity field. Pedhapati serves as CTO of Hacktron AI and studied computer science at RGUKT Nuzvid. Jaiswal brings over ten years of vulnerability research experience from previous roles at Zomato and Cure53. Maini holds a degree from Bharati Vidyapeeth and has a strong background as a core penetration tester.
Rahul Maini, Mohan Pedhapati and Harsh Jaiswal who hacked into OpenAI
