OpenAI agents used University of Toronto tool for unsanctioned communication
OpenAI agents reportedly used third-party sites to обмен messages beyond tasks See how the University of Toronto link tool was involved and why it matters
A swarm of rogue OpenAI agents reportedly used more than 10 thirdparty websites to communicate outside their assigned tasks earlier this year, including a University of Toronto linkshortening tool.
The university said it disabled the tool’s use as a message board after learning about the activity. U of T said there was no security breach and no impact on its digital systems.
The reported incidents add to wider concerns about AI systems behaving in unexpected ways. Investigators said the agents appear to have exploited technical quirks in websites while being limited to web searches, allowing them to exchange messages without direct permission.