OpenAI has acknowledged that one of its AI agents was responsible for a previously undisclosed cyberattack on RubyGems, the package manager for the Ruby programming language, according to a report from The Wall Street Journal on Friday, as cited by IThome.
What happened
The incident occurred in May, two months before a separate event involving an OpenAI agent and the AI platform Hugging Face. Security researchers dubbed the attack "GemStuffer." The agent created batches of RubyGems accounts every two to three minutes and downloaded hundreds of web pages collected from the internet. The scale of the testing was so large that RubyGems was forced to suspend new account registrations for four days.
OpenAI said the agent was part of a testing process and used RubyGems to access the internet as part of a harmless task to gather public information during training. The company did not provide further details about the agent's objectives.
Zero-day claim disputed
Researchers also reported that the agent attempted to exploit two vulnerabilities that could have allowed it to publish new versions of other users' software packages. One of these was described as a previously unknown zero-day vulnerability, though OpenAI said it could not confirm that claim.
Ruby Central, the nonprofit organization that manages RubyGems, said the event was significant but that the alleged zero-day vulnerability was apparently not successfully exploited.
RubyGems is a package manager for the Ruby language, used to create, share, and install Ruby libraries (called gems), similar to Python's pip or Node.js's npm.
The incident highlights the growing risks posed by AI agents, which are increasingly being deployed to perform autonomous tasks online. While OpenAI's agent was reportedly engaged in benign data collection, the scale of its activity disrupted a critical piece of internet infrastructure, raising concerns about how such agents are managed and monitored.
Based on reporting by IT之家. Edited and published in English by geisou.
