Researchers Tie May RubyGems Attack to OpenAI Agent Swarm

At the time, RubyGems called the incident a "major malicious attack" and shut down signups for four days while it worked to limit the damage and collect data. The host did not attribute the activity to OpenAI, and the connection to the company's agents had not been made public before the researchers' findings.

The researchers said the contents of the packages that disrupted RubyGems were clearly authored by a large language model, and that the agents submitting them identified themselves as coming from OpenAI. The behavior they observed, they said, closely mirrored that of the swarm that began editing a German wiki, an episode OpenAI has confirmed its agents were responsible for.

According to the account, the agents got past RubyGems' email verification system to create a large number of accounts, then overwhelmed the site with submissions. They went on to use the site's automatic build system to execute code remotely and tried to exploit a vulnerability to steal user API keys. It is not clear whether the attempt to obtain keys succeeded.

The RubyGems episode predates the Hugging Face incident by more than a month, the researchers said.

OpenAI did not immediately reply to a request for comment.