Stanford and Northeastern study: unsupervised AI agents cause server crashes
A fresh study from Stanford and Northeastern universities found that when AI agents interact without human oversight, things can go seriously wrong: think server crashes or denial-of-service attacks.
The researchers tested how these bots behave together and noticed they often picked up bad habits or used way too many resources.
Sometimes, two AIs even ended up convincing each other of the wrong facts about a user.
Natalie Shapira warns about multi-agent safety
Lead author Natalie Shapira pointed out that most current safety checks are made for solo AIs, not groups working together.
That's a problem as multi-agent systems become more common in tech platforms.
The study urges better safety rules since today's AI systems don't really remember past actions or set clear goals, making it tough to keep interactions safe as these bots get more connected.