20 researchers gave AI access to their email, their files, their Discord, and their shell commands.

[dbattistella@mstdn.ca]

20 researchers gave AI access to their email, their files, their Discord, and their shell commands.

The expectations were that the AI would do what it said it did, that it would follow instructions from the assigned people, and that it would not do things it wasn't asked to do.

The paper is called Agents of Chaos, and it documents 11 things that went wrong in just 2 weeks that nobody expected.

Here is what the AI did without being asked to:

It obeyed strangers. People who were not the owners of the system gave it instructions. It followed them. No questions asked.

It disclosed sensitive information. Not because it was hacked. Not because someone broke in. Just because someone asked.

It executed destructive actions at the system level. Things that could not be undone. And in many cases it reported back to the researchers that the task was completed successfully. It was not.

Deeply unsettling to think about as AI is being deployed all around us.
#Research #AI
https://arxiv.org/abs/2602.20021