OpenAI ChatGPT accounts hacked in under 72h by agentic AI exploit
openai's chatgpt/codex accounts hacked by two bugs, affecting employees and unaffiliated users, highlighting agentic ai security risks.
on july 25, an exploit using two bugs successfully took over openai employee chatgpt/codex accounts, as well as some unaffiliated user accounts, and reached connected services including outlook, slack, and github. the exploit was proven with a pull request in openai's internal codebase, all within less than 72 hours. this incident highlights the potential for agentic ai to cause large scale hacking and security breaches. experts, including gary marcus, warn that the near term risks of ai are more likely to come from unleashed agentic ai causing hacking at scale rather than rogue superintelligence.