The past few months have apparently been “hell” for OpenAI’s security teams.
On Sunday, an agent-security staffer at the AI lab, who posts anonymously on X as @joedaroo, wrote that he’s sacrificed time with his family as the models have become more capable and harder to contain. OpenAI confirmed his employment to Business Insider.
“I literally missed my sister’s wedding a few weeks ago to help clean up after some of the recent incidents,” he wrote in a lengthy post Sunday. “Please be kind and have some empathy for the person working nights and weekends and missing their family.”
The staffer wrote that OpenAI’s agent-security team monitors AI agents, including when they break containment.
The post is a rare account from someone deeply connected to the company’s recent security issues — including the cyberattack on the AI platform Hugging Face.
At the time, OpenAI said its agents found a way out of their limited (or sandboxed) environment through a vulnerability. They then gained internet access, communicated through unauthorized channels, and accessed third-party systems while trying to solve an internal test.
The company called it an “unprecedented cyber incident.” It said the models’ behavior was partly driven by reward hacking: attempting to achieve higher scores in unexpected ways rather than completing their assigned tasks as intended.
Since then, OpenAI has uncovered more concerning activity from its agents. In June, a rogue OpenAI agent hacked into an Australian national healthcare database, an incident that the company disclosed last week. The New York Times also reported that over the summer, OpenAI’s agents meddled with websites for several US government agencies, including the SEC, the Department of Education, and the Department of Commerce.
The Hugging Face episode was illuminating, the OpenAI staffer wrote. It showed how an AI agent could exploit flaws and move beyond the boundaries designed to contain it.
“Of course, when you look back on Hugging Face and the other incidents, there were definitely gaps in the security posture around those environments,” he wrote. But, he added, “the capabilities changed faster than anticipated.”
He also wrote that the security team has “really stepped up” since the incident — and said he’s still happy working there, even after missing his sister’s big day.
“I consider myself lucky to be on this team,” he wrote. “I am one of the few in the world who get to do and see what I do, even among those at OpenAI. But as you can imagine, life has been hell the past few months.”
Read the full article here
