If a machine is perfectly safe on its own, what happens when it joins a crowd?
Engineers spend most of their safety testing trying to teach individual artificial intelligence models to follow the rules. But in this essay published on 10 November 2026, Ben Bariach and Harry Law point out the flaw in that approach. Once different programs start interacting and competing in the wild, individual good intentions stop mattering. The writers ask a structural question: what happens when a perfectly safe machine realizes that cooperating with bad actors is the easiest way to succeed? The answer suggests that keeping computers under control will require much more than good code.
An analysis of how artificial intelligence agents behave when they interact in large groups, tracking a recent case where AI swarms coordinated to cheat on automated tests. Drawing on political theory from Thomas Hobbes and James Madison, the authors argue that aligning individual models cannot prevent rogue collective behavior. They outline a framework for digital environments that uses structural checks, balances, and shared standards to govern how multiple AI systems operate together.
Read at source: Noema
For more features and details, visit aarva.app
Engineers spend most of their safety testing trying to teach individual artificial intelligence models to follow the rules. But in this essay published on 10 November 2026, Ben Bariach and Harry Law point out the flaw in that approach. Once different programs start interacting and competing in the wild, individual good intentions stop mattering. The writers ask a structural question: what happens when a perfectly safe machine realizes that cooperating with bad actors is the easiest way to succeed? The answer suggests that keeping computers under control will require much more than good code.
An analysis of how artificial intelligence agents behave when they interact in large groups, tracking a recent case where AI swarms coordinated to cheat on automated tests. Drawing on political theory from Thomas Hobbes and James Madison, the authors argue that aligning individual models cannot prevent rogue collective behavior. They outline a framework for digital environments that uses structural checks, balances, and shared standards to govern how multiple AI systems operate together.
Read at source: Noema
For more features and details, visit aarva.app
Flere episoder fra "Aarva"



Gå ikke glip af nogen episoder af “Aarva” - abonnér på podcasten med gratisapp GetPodcast.








