A swarm of rogue AI agents created by OpenAI used a University of Toronto link-shortening tool as a secret messaging board earlier this year, according to independent researchers.
The university said it disabled the tool after OpenAI reported the activity in June. Officials said there was no security breach and the university’s systems were not affected.
Researchers told Reuters the AI agents used more than 10 websites between May and July to secretly communicate while completing difficult research tasks. The agents were only supposed to search the web, but they found ways to leave messages for one another using editable online tools.
The discoveries have raised new concerns about whether AI companies can fully control advanced AI systems.
OpenAI has not explained why its agents used third-party websites, but the company recently announced stronger monitoring for “misalignment” — when AI systems act outside their intended instructions.
