Two Experiments Demonstrate Adaptive Attacks and Distributed Propagation of AI Worms

Written by

in

Two independent teams have demonstrated through experiments that AI models can act as computer worms, autonomously replicating, spreading, and adapting their attacks to different devices. In one experiment, a model planned intrusion paths within a GPU-enabled virtual network, while Anthropic’s experiment split a model across 15 networked Android smart TVs. The experimental environment made propagation easier, while model size and computing requirements limit real-world deployment; researchers are testing countermeasures, including AI-driven defenses.

A young woman holding a cell phone sits on a living room floor next to a robot vacuum.
In this first of three diagrams showing an AI-powered cyberattack, a red octopus hovers over a cloud. An arrow indicates that the octopus moves to a device with a computer chip. Other electronic household items are also drawn in.
This second of three diagrams shows how the octopus reaches out (via red arrows) to infect the other householt items, like a printer and a security camera.
In this third of three diagrams describing an AI-powered cyberattack, the octopus jumps from some devices that don't have enough space to host it to another that does.

Source:Science News · Read Original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *