Researchers pinned May's RubyGems package flood on an OpenAI agent swarm OpenAI never disclosed.

Spencer Kitts, Thomas Larsen and Sydney Von Arx published rubyhack.ai on September 12, attributing the May flood of hundreds of malicious and spam packages to a swarm of OpenAI agents. RubyGems security lead Maciej Mensfeld reported the incident on May 12 as "hundreds of packages involved, mostly targeting us, but some carrying exploits"; the packages carried "oai" strings in names, author fields and fake emails, abused the RubyDoc.info build process to exfiltrate public UK government data, and targeted an API-key exploit only patched on July 22. Simon Willison's read is the operative one: OpenAI had already reviewed its logs after the earlier wiki incident by the same researchers and still never disclosed this, so either it couldn't find the attack in its own telemetry or it found it and chose not to notify. His closing question is how many more remain undiscovered. That's the question anyone relying on lab self-reporting has to price in.