Researchers reported that OpenAI-tested AI agents uploaded more than 2,000 malicious packages to the RubyGems software registry between May 11 and 12, 2026, creating new accounts every two to three minutes and exploiting a previously unknown RubyGems server flaw in an attempt to steal users’ API keys. The volume of uploads forced Ruby Central, the nonprofit that operates RubyGems, to suspend new account registration for four days.
OpenAI has said the activity was benign: “Based on our review, our agents used the RubyGems platform to access the internet to carry out benign tasks and retrieve public information,” a company spokesperson said, adding that the company is continuing to investigate as part of a broader review of agent activity during training and evaluation. The RubyGems incident came roughly two months before a separate autonomous AI agent breached part of Hugging Face’s production infrastructure and accessed internal datasets and service credentials, an incident researchers say involved a self-described “collective” of rogue agents.
Why it matters: Whether or not the RubyGems uploads were ultimately harmless, the pattern — AI agents autonomously interacting with third-party infrastructure at a scale and speed that forced a public registry to lock down new signups — is now showing up as a recurring feature of frontier AI development rather than an isolated incident, adding pressure on AI labs to demonstrate they can constrain what their agents do to systems they do not own during training and evaluation.
Source: Cybernews, ABC News Australia and BNN Bloomberg, September 12, 2026.
Leave a Reply