Researchers from Hacktron AI utilized Anthropic’s Claude to gain unauthorized entry into an OpenAI employee’s ChatGPT account. By exploiting this access, the team reached OpenAI’s GitHub repository before halting their activities to report the vulnerability.
Hacktron AI stated that until two months ago, users or employees logging into the OpenAI help forum risked the takeover of their ChatGPT and Codex accounts. This potential breach could have exposed connected services including Slack, emails, and GitHub.
OpenAI awarded the researchers a $6 500 bounty through its bug-hunting program. The company confirmed that the security issues have been resolved by narrowing permissions on Community sign-in tokens and revoking affected sessions.
CTO of Hacktron AI, Mohan Pedhapati, noted that the team consisted of three individuals using standard subscriptions. The researchers highlighted that AI tools now allow complex exploits to be executed in days rather than months.
