Security Researchers Breach OpenAI Using Claude Opus 5 in 72-Hour Challenge

A team of three security researchers using Anthropic's Claude Opus 5 successfully penetrated OpenAI's internal systems by chaining two previously unknown vulnerabilities in less than 72 hours, exposing how rapidly advancing AI capabilities can enable sophisticated cyberattacks. The researchers—Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini operating under the collective name Hacktron—gained access to employee ChatGPT accounts and OpenAI's internal GitHub repository through the company's community forum during an authorized bug bounty engagement, demonstrating both the promise and peril of frontier AI models.