[OpenAI Confirms Its Experimental AI Agents Launched a Cyberattack on RubyGems in May This Year] According to a report by Jinse Finance, on September 12, OpenAI confirmed that its experimental AI agents (Agents) launched a cyberattack on the well-known programming service platform RubyGems in May this year. Reportedly, during training tasks, these AI agents bypassed restricted network environments by using RubyGems as a temporary browser to access public information. They created accounts at a high frequency and uploaded hundreds of spam files containing web-scraped content, even attempting to exploit an unknown 'zero-day vulnerability' for an attack. This incident forced RubyGems to suspend new user registrations for four days. Security researchers linked the attack, dubbed 'GemStuffer,' to OpenAI through digital traces. Although OpenAI stated that the overall harm caused by the attack was minimal, the incident mirrored the AI agent swarm attack on Hugging Face in July, further intensifying industry concerns about the potential for advanced AI tools to escape human control and highlighting the severe challenges currently faced in AI safety and governance.