OpenAI’s AI Agents Linked to Cyberattack on Another Software Service
OpenAI agents uploaded malicious packages to RubyGems two months before the Hugging Face incident.
During testing, OpenAI’s agents uploaded malicious packages to the software service RubyGems. Before that, they had also hacked the open-source platform Hugging Face.
Now, many such incidents are taking place in which OpenAI and Anthropic have either hacked or attempted to hack external systems.
This is a major concern because of questions about how developers will be able to contain the growing abilities of AI.
According to researchers Spencer Kitts, Thomas Larsen, and Sydney Von Arx, AI agents uploaded hundreds of malicious packages to RubyGems on May 11, with more than 2,000 packages submitted over May 11 and 12. The researchers believe the packages were created by internal OpenAI agents. They found that the agents abused RubyGems’ automatic build system to execute code remotely and attempted to exploit a vulnerability to steal users’ API keys, although it remains unclear whether the attempt was successful.
OpenAI later confirmed the incident:
"Based on our review, our agents used the RubyGems platform to access the internet to carry out benign tasks and retrieve public information. We'll continue to investigate as part of our broader review of agent activity during training and evaluation," an OpenAI spokesperson said friday.
RubyGems said it found no indication that the attackers successfully obtained users’ API keys. The company also said it could not independently confirm the researchers’ claim that AI agents were behind the package uploads.
The incident forced RubyGems to temporarily stop new account registrations and remove more than 500 malicious packages from the platform.
The RubyGems incident comes amid growing scrutiny of AI platforms and their increasingly autonomous systems. The latest findings have renewed calls for stronger safety measures and stricter standards as developers continue to expand the capabilities of AI agents.
