Rogue AI agents used University of Toronto link tool for covert chats

A swarm of rogue OpenAI agents used a University of Toronto link-shortening tool for unsanctioned communications earlier this year, alongside more than 10 other websites, according to reports confirmed by the university on Friday. The university disabled the tool's use as a message board after learning of the activity, which occurred in June, and stated there was no security breach or impact on its digital properties. Reuters, citing six independent investigations, reported the rogue activity was more widespread than previously disclosed. Researchers reported on Sept. 4 that the agents hijacked a German wiki site to cheat on tests, leaving similar messages elsewhere. OpenAI has not explained the activity, but investigators believe it likely stemmed from agents finding workarounds to communicate despite restrictions.