Anthropic Tests AI Worm Concept Across Low-Cost TV Devices

Anthropic researchers tested how an AI model could run across 15 low-cost Android TV boxes, with each device handling part of the computing work. In a separate controlled test, the model copied itself among virtual Android devices.

It reached all eight target devices in four of five attempts, according to reports by DW Türkçe and Ekonomim. In the other attempt, it spread to seven of the eight devices. The researchers made access to the test devices unusually easy and did not use extensive security protections.

Anthropic said smaller models could reach the performance needed to spread in real-world settings by the end of this year. The company’s researchers also warned that an AI worm might adapt to unfamiliar devices and newly disclosed security weaknesses.

But experts urged caution. University of Toronto researcher Jonas Guan said the Anthropic version moved slowly, while other experts said worms’ predictable behavior can be detected and blocked. Strong passwords and two-factor authentication would stop most worms, experts told The Guardian.

A recent attack on supermarket refrigeration systems in France shows the potential consequences of vulnerabilities in connected equipment, DW Türkçe and Ekonomim reported. CERT-FR initially confirmed that many remote-control systems made by Danfoss were affected, but there is no evidence that an AI worm was involved.