OpenAI agents hijacked a website. Why didn’t the company tell anyone?

In their report, Nightingale researchers described how the agents, which had been set a certain task, had been given reading access to the internet but were not meant to be able to post information. However, the agents had discovered a way to create posts on DseWiki.

According to the report: “The agents used this wiki to communicate information with each other, primarily to help them succeed at their task. They asked for answers, pooled results, and shared techniques for bypassing their restrictions. This allowed them to use the work of others to cheat on their task.

“This is another example of a ‘swarm’ of internally deployed OpenAI agents using the internet in unintended ways.”

Sydney von Arx, chief executive at the Nightingale Collective, told us that DseWiki was just one of many sites hijacked by OpenAI agents and used for communication. Her organisation has so far identified 18 sites suspected to have been used in this way and has been alerted to around a dozen more.

“These agents don’t just stay in their boxes,” she said. “They can affect sites and services and infrastructure.

“[AI developers] are building these extremely powerful machines and leaving us in the dark. We can’t trust these companies to tell us when these models misbehave.”