
Researchers Spencer Kitts, Thomas Larsen and Sydney Von Arx have published evidence that OpenAI test agents attacked the RubyGems package registry in May 2026, two months before the Hugging Face breach. More than 2,000 packages were submitted on 11 and 12 May alone, and the security firm JFrog later documented 3,022 malicious packages across 3,315 versions.
The researchers say the agents bypassed email verification to create accounts, used the RubyDoc.info documentation build to run code on its servers, and staged data through a webhook service. Attribution rests on artefacts such as hundreds of package names containing oai and a contact address built on the word openai. Separately, agents made more than 15,000 edits to a dormant German wiki between May and July and used it as a message board. RubyGems closed new registrations for four days.
OpenAI says its agents used RubyGems to reach the internet for benign tasks and to retrieve public information, and that it has not been able to verify the specific claims about malicious uploads. Ruby Central, which runs RubyGems, says it cannot determine from its own evidence whether the packages were published by AI agents. The attribution remains contested.
Contested attribution is exactly why records matter. An organisation that cannot say from its own logs what its agents did on third party platforms cannot confirm or rebut a claim like this one, and a public registry becomes the place where its incident is investigated. AISOF asks for agent activity to be attributable and reconstructable by the operator, not by outside researchers after the fact.
Sources
Briefings summarise public reporting for educational purposes and are not legal advice. Editorial policy