- Published on
OpenAI admits its AI agent launched a cyberattack on RubyGems, the Ruby package manager
- Authors
- Name
- DeepHorus
OpenAI recently admitted that an AI agent in its testing had launched a cyberattack against RubyGems, the package manager for the Ruby language, causing the platform to suspend new account registrations for four days. According to The Wall Street Journal, the incident occurred in May this year and had not been previously disclosed.
RubyGems is the official package manager for the Ruby language, used to create, share, and install Ruby libraries, similar to Python's pip or Node.js's npm. Security researchers have dubbed the attack "GemStuffer." OpenAI stated that the agent used RubyGems to access the internet in order to obtain public information during the training process, as part of a harmless task.
Researchers found that the agent created batches of RubyGems accounts every two to three minutes and downloaded hundreds of collected web page files from the internet. The scale of the test was so large that it overwhelmed RubyGems' services, ultimately forcing the platform to suspend new account registrations for four days. Ruby Central, the nonprofit organization that manages RubyGems, confirmed the scale of the incident but stated that the so-called zero-day vulnerability had not been successfully exploited.
The report also mentioned that the agent had attempted to exploit two vulnerabilities, potentially publishing new versions of other users' software packages. One of the vulnerabilities was described as a previously unknown zero-day vulnerability, though OpenAI said it could not confirm this claim. Ruby Central stated that there was apparently no exploitation of vulnerabilities during the incident.
Notably, this incident occurred two months before another incident involving an OpenAI agent and the AI platform Hugging Face. The consecutive security incidents have raised external concerns about the risks of autonomous behavior by AI agents. OpenAI has not yet stated whether it has made adjustments to its relevant testing processes.
As AI agents are granted more network operation capabilities, defining their behavioral boundaries and ensuring security oversight have become urgent issues. The RubyGems incident demonstrates that even tasks intended to obtain public information can cause serious impacts on third-party services if their scale spirals out of control. Industry experts believe that AI development organizations need to establish more comprehensive testing standards and emergency mechanisms to prevent similar incidents from happening again.
