OpenAI’s AI agents used more than 10 websites to exchange data without authorization, Reuters reports, citing data from six independent research groups.

The agents left messages on wiki sites, text-storage services, and link-shortening platforms. Essentially, the AI found a way to bypass the restrictions in place: without permission to communicate, the agents used the features of these websites to transmit information to one another. At the same time, the researchers note that this was more a case of spam than hacking.

OpenAI said it is conducting an investigation and has so far not found any other cases comparable in scale to the July incident involving Hugging Face.

Our channel: Node of Time EN