By Kenrick Cai
July 29 (Reuters) - New revelations have been surfacing about the AI agent that breached an OpenAI testing environment to hack into AI library Hugging Face.
My colleagues Raphael Satter, Deepa Seetharaman and I exclusively reported last week that the hacking spree had been going on for days before OpenAI even realized its agent was behind the hack.
On Tuesday, Hugging Face published a detailed technical timeline that alluded to the agent having broken into misconfigured infrastructure hosted by an unnamed third-party provider. OpenAI published an update on its investigation, finding that the agent broke into four accounts across four separate services.
Reuters confirmed that the third-party provider alluded to by both firms was Modal, a New York-based cloud infrastructure provider. The startup’s CTO told us that the company itself was not hacked, but the agent exploited a customer’s vulnerable code which had been publicly accessible.
In the fallout, support has consolidated around mechanisms to counteract future incidents. Last week, Nvidia, Microsoft and other tech heavyweights made a case to U.S. lawmakers in favor of open-weight models that users can download, run and customize. (Notably, Hugging Face previously detailed that it turned to a Chinese open-weight model to fix the breach, after having been blocked from using proprietary models due to their baked-in guardrails.)
On Tuesday, more than 1,100 employees from top tech firms signed a statement urging the U.S. government to support an international effort to manage the pace of AI development. Many of the signatories work at OpenAI, including chief scientist Jakub Pachocki, chief research officer Mark Chen and co-founder Wojciech Zaremba.
One of the signatories pinpointed the Hugging Face hack as “a clear and undeniable warning sign that we aren’t yet prepared to handle AI systems that demonstrate capabilities beyond those of our smartest people.”
Hugging Face CEO Clement Delangue told Reuters that the shift in discourse around AI development in the aftermath of the hack has left him feeling more optimistic, pointing to support for open-weight models.
“Now that everyone expressed their public support for open weights in America, I hope it will drive more companies to actually share more of their research, models and datasets,” Delangue said.
In this week’s issue, we look at the tech earnings bonanza, and what Google parent Alphabet’s results last week may portend. Scroll on.
XLK•
