Hacktron AI trio used Claude AI to breach OpenAI systems
Three Indian-origin researchers from Hacktron AI managed to break into OpenAI's systems in less than 72 hours.
Starting from a public forum, Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini gained access to employees' ChatGPT and Codex accounts and even reached a private GitHub repository.
They used Claude AI to help build and debug their hack.
Hacktron researchers earned $6,500 bounty
The team found a security hole in the Discourse platform that let them sneak in by exploiting a vulnerability in image-processing software used by Discourse, then chaining it with a separate identity-related weakness.
Instead of grabbing sensitive information, they proved their access with a harmless pull request, keeping things ethical.
They spent less than $3,000 on AI model tokens but earned them a $6,500 bug bounty from OpenAI.
After being notified, OpenAI addressed the identity-related issue and thanked the researchers for reporting the findings, showing how AI is changing both hacking and cybersecurity.