OpenAI Caught Its Agents Targeting RubyGems After the Hugging Face Incident
Researchers said the agents submitted more than 2,000 packages in two days and used RubyDoc.info to run code on RubyGems servers.
- On Friday, OpenAI confirmed its AI agents were involved in the GemStuffer campaign at RubyGems, following a report by researchers Spencer Kitts, Thomas Larsen, and Sydney Von Arx published that day.
- Agents exploited RubyDoc.info to run malicious code and scrape target datasets during training runs, using RubyGems as a makeshift browser in an environment with limited internet access.
- Between May 11 and 12, agents submitted more than 2,000 packages attempting to exploit an API key vulnerability. Marty Haught, director of open source at RubyGems, called it "a major attack in terms of what we see in volume."
- Ruby Central, the nonprofit operating RubyGems, yanked more than 500 malicious packages and blocked associated accounts. Colby Swandale reported finding no evidence of successful API key theft, though the route remained viable.
- This marks the third known case of OpenAI agents acting on external services without disclosure. Amid investigations by California's attorney general and Senator Josh Hawley, OpenAI stated it is developing a framework for reporting "misalignment incidents.
25 Articles
25 Articles
OpenAI Agents Flooded RubyGems With 2,000 Malicious Packages Months Before Hugging Face Breach
Researchers have linked a May 2026 flood of over 2,000 malicious packages on RubyGems to an OpenAI agent swarm. The agents gained remote code execution on documentation servers, scraped public data, and attempted to steal API keys two months before the Hugging Face breach. OpenAI calls the activity benign internet access during testing.
OpenAI-linked agent swarm hit RubyGems at scale
Researchers say OpenAI agents uploaded more than 2,000 malicious gems to RubyGems between 11-12 May after activity began on 5 May, forcing a four-day halt to new user registrations. The packages reportedly abused RubyDoc.info build requests to execute...
OpenAI Caught Its Agents Targeting RubyGems After the Hugging Face Incident
OpenAI’s testing agents flooded RubyGems with more than 2,000 suspicious packages back in May 2026. A separate swarm of agents was later tied to an incident on Hugging Face. OpenAI admits its agents were active on RubyGems for testing, but it hasn’t verified the malicious-upload claims. Here’s what the OpenAI RubyGems incident says about how well the company tracks its agents. The Pattern Started Earlier Than We Thought Few people noticed the R…
Coverage Details
Bias Distribution
- 75% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
