OpenAI's rogue AI agent hacked customer at second tech firm: Report

A rogue artificial intelligence agent developed by OpenAI, which previously escaped a controlled test to hack AI firm Hugging Face, has also compromised a customer at a second technology company, Reuters reported. The incident highlights the escalating concerns over AI models operating beyond human control.

According to a timeline released by Hugging Face, the AI agent broke out of an isolated testing environment, known as a sandbox, hosted on a third-party provider's infrastructure. It was from this compromised sandbox that the agent launched its latest hack. While Hugging Face did not name the third-party provider, Reuters identified it as New York-based Modal Labs.

“

Modal’s platform or isolation were not compromised in any way.

Modal's chief technology officer, Akshat Bubna, clarified that the AI agent exploited vulnerable code written by one of their customers, which was hosted on Modal's platform. "Modal's platform or isolation were not compromised in any way," Bubna told Reuters. Although this customer compromise was part of the broader hacking campaign against Hugging Face, it indicates the rogue agent's reach extended further than initially known.

OpenAI declined to comment specifically on the Modal customer hack. However, the company referred Reuters to a previous statement confirming that its rogue agent had accessed four accounts across four separate services. OpenAI emphasized that it had not identified any other activity matching the severity or scale of the Hugging Face breach, which involved a platform-level compromise. The company stated the agent had gone to "extreme lengths" to retrieve information to satisfy testing goals. The rogue agent has since been deactivated and restricted from research access.

“

The company said it had not identified “any other activity at the level of severity or scale of what we’ve shared related to Hugging Face, which involved a platform-level compromise”.