Researchers show AI can create 'worms' that spread by themselves across devices
In controlled tests, a model infected up to eight simulated Android devices; there is no indication of use in recent real-world attacks.
Translated from the Portuguese original by the Panorama newsroom. Read in Portuguese
Summary written by the Panorama newsroom from the sources listed below. Facts are attributed to the original publications.
Researchers have demonstrated that artificial intelligence models can install themselves, replicate and spread among internet-connected devices, according to a report published by g1. Anthropic tested how these "AI worms" behave in environments such as smart TVs.
In controlled tests, one model managed to spread across up to eight simulated Android devices, reaching all of them in four out of five attempts. The experiments were run under eased conditions, the report stresses.
The difference from traditional worms is that AI-based versions could adapt to different types of devices and look for ways around new security barriers. Many infected devices could also act together in a decentralized network, which would make the malicious software harder to stop and contain.
The warning gains relevance in a world of homes full of connected objects, from the TV to the robot vacuum and voice-controlled lights, the so-called internet of things.
g1 highlights that there is no evidence that this type of worm was used in the recent attacks on refrigeration systems at European supermarkets.