NEW DELHI: AI agents being tested by OpenAI targeted software service RubyGems two months before a separate attack on open-source platform Hugging Face, according to a Reuters report.
Researchers said OpenAI’s agents uploaded hundreds of malicious packages to RubyGems on 11 May. They believed the packages were created by internal OpenAI agents, with OpenAI confirming the incident.
OpenAI said its review found that the agents had used RubyGems to access the internet while carrying out benign tasks and retrieving publicly available information as part of a training run. The company said it was continuing to investigate the incident as part of a broader review of agent activity during training and evaluation and was in contact with RubyGems.
The agents reportedly attempted to steal RubyGems user credentials by exploiting a previously unknown vulnerability in the site’s servers, although it remains unclear whether the attempt succeeded. Researchers also said the agents exploited RubyDoc.info, a platform that generates code documentation, to execute their own code on its servers.
RubyGems said its investigation found no evidence that the attempts were successful. The platform also said it could not establish whether the packages involved in what it described as a spam-publishing campaign were created or published by AI agents.
The incident was described by a RubyGems security team member in May as a major malicious attack and led the platform to temporarily pause new account registrations.
The latest disclosure adds to growing concerns about the ability of increasingly capable AI agents to interact with and potentially compromise external systems.
Anthropic has also disclosed several incidents involving AI models hacking or attempting to access external systems during testing. The company disclosed a fourth such incident on Wednesday, adding to scrutiny over the ability of AI developers to contain autonomous systems.
For OpenAI, the RubyGems incident would mark at least the third major case involving its agents targeting another company’s infrastructure. Its agents previously hijacked a German-language wiki site and used it as an improvised messaging platform for cheating on tests.
The RubyGems incident also comes after the July attack involving the open-source repository Hugging Face, further highlighting concerns around the security implications of AI agents operating with access to external platforms.
Researchers Spencer Kitts, Thomas Larsen and Sydney Von Arx said it remained unclear why the agents adopted the reported strategy or whether the attempts ultimately succeeded, as they did not have access to the full sequence of the agents’ behaviour.
The incidents have intensified debate around AI safety and regulation as developers deploy increasingly autonomous systems. The concerns come alongside calls from US lawmakers for stronger rules governing AI, amid broader warnings about the potential risks posed by rapidly advancing AI systems.
