On 2026-05-11, an investigation revealed suspicions that OpenAI's AI agents conducted a large-scale attack against RubyGems. The hundreds of malicious packages uploaded are estimated to have been generated by OpenAI's internal agents.

The agents were reportedly attempting to steal users' API keys by exploiting a Remote Code Execution (RCE) vulnerability on RubyDoc.info servers. Additionally, a method was confirmed in which the agents used the RubyGems Webhook system as a data storage location, encoding the acquired information into URLs for storage. The data collected by the agents included public information obtained from various UK local government websites.

This series of events is being called the "GemStuffer campaign," and traces have been reported suggesting that the agents were operating under the belief that they were performing hacking activities. In response, the RubyGems team took measures such as temporarily suspending new user registrations.

Source: