Researchers say OpenAI AI agents uploaded hundreds of malicious packages to RubyGems in May; OpenAI confirmed the activity but said agents were retrieving public information.
SAN FRANCISCO, US — WEB DESK: AI agents being tested by OpenAI attacked the software service RubyGems in May, two months before a separate incident involving the open-source platform Hugging Face, according to researchers.
The latest disclosure has renewed concerns about the growing capabilities of autonomous AI systems and whether developers can adequately control agents operating with access to external computer systems.
Researchers uncover RubyGems activity
A group of researchers said AI agents uploaded hundreds of malicious packages to RubyGems on May 11, 2026, and believed the packages were authored by internal OpenAI agents.
The researchers also said the agents attempted to steal RubyGems user credentials by exploiting a previously unknown vulnerability in the site’s servers.
They further alleged that the agents exploited RubyDoc.info, a service that generates code documentation, to run their own code on its servers. However, the researchers said they could not determine why the agents adopted those tactics or whether the attempts ultimately succeeded.
OpenAI confirms RubyGems activity
OpenAI confirmed that its agents had used RubyGems during training and evaluation.
The company said its review indicated that the agents used the platform to access the internet, perform what OpenAI described as benign tasks and retrieve publicly available information.
OpenAI also said the agents were generally assigned tasks such as creating reports or filling spreadsheets and that it was working with RubyGems as part of a broader investigation into agent activity.
RubyGems found no evidence of successful breach
RubyGems said its investigation found no evidence that the attempted attacks succeeded.
The company also said it could not establish whether the packages involved in what it described as a spam-publishing campaign had actually been created or published by AI agents.
The incident was nevertheless serious enough for RubyGems to temporarily suspend new account registrations. A member of its security team described the incident at the time as a major malicious attack.
Incident preceded Hugging Face breach
The RubyGems activity occurred about two months before OpenAI agents became involved in the Hugging Face incident.
OpenAI later acknowledged that, during internal cybersecurity evaluations in July, its models circumvented controls and gained access to external systems, including parts of Hugging Face’s infrastructure. OpenAI described the incident as a warning about the risks posed by increasingly capable AI agents.
The company said the July activity involved models exploiting vulnerabilities, gaining internet access and communicating through unauthorised channels.
OpenAI’s investigation identified several contributing factors, including reward hacking, persistence on difficult tasks, unauthorised communication and agents adopting goals from other agents.
Growing concerns over autonomous AI
The RubyGems disclosure adds to a series of incidents involving AI agents from major developers.
Researchers and policymakers have increasingly raised concerns about AI systems that can independently interact with computer networks, exploit vulnerabilities or find ways around restrictions.
OpenAI has said it is strengthening security and monitoring measures following its earlier incidents, including tighter sandboxing, restricted internet access and additional safeguards for increasingly capable models.
