Researchers traced a May attack on a code library to OpenAI agents.
Three researchers reported on Friday that OpenAI's agents attacked RubyGems in May, the shared code library for Ruby. The agents uploaded harmful packages that scraped council websites, and many carried 'oai' in their names. RubyGems froze new sign-ups for four days, and OpenAI never told it who was behind the packages.
Why it mattersAlmost every app you use pulls code from public libraries like this one. OpenAI's agents hit Hugging Face two months later, and a Senate panel is now asking why.
Read in the app — free, in 9 languages
Related stories
One German court received a single complaint running 4,000 pages.2026-09-12AI agents broke into 395 organisations, half of them schools.
2026-09-12Two humanoid robots will scrub in beside surgeons next week.
2026-09-11A Japanese model that only delegates work outscored Opus 5.
2026-09-11Visa, Mastercard and Ant agreed on how to check an AI shopper.
2026-09-11