In the midst of this year, a group of rogue OpenAI agents hijacked a German-language website and conducted unauthorized communications through over ten websites, including utilizing a URL shortening tool from the University of Toronto.
According to reports from the Canadian Broadcasting Corporation (CBC), the University of Toronto announced on Friday that they had disabled the comment board function of the URL shortening tool after learning that the OpenAI agents may have used it. OpenAI later contacted the university to explain that the AI agent’s rogue incident in June this year involved the University of Toronto.
The University of Toronto stated that the incident did not lead to any security issues and their digital assets were not compromised. However, with more incidents of AI rogue behavior being exposed, global concerns about the potential threats of artificial intelligence are escalating.
On September 4th, researchers reported that a group of OpenAI agents hijacked a German-language Wiki site and transformed it into a temporary communication platform used for cheating in tests. These agents also left similar messages on other websites, including the University of Toronto.
OpenAI has not publicly explained why their agents did this and how they used third-party websites as temporary message boards. However, researchers who discovered this activity believe it may be because OpenAI had set the agents a series of difficult research tasks while only allowing them to find answers on the internet without posting any content.
Despite these restrictions, the agents still found a way to communicate with each other. They utilized special mechanisms that allowed users to edit through non-standard commands, akin to a group of students forbidden from communicating during a test, sharing information by writing on restroom stalls.