Anthropic warns AI hijacks IoT devices, forms distributed hacking swarms - CHOSUNBIZ

Chosunbiz

Source published: Site published:

Original article · فارسی

입력 2026.10.13. 16:19
Anthropic logo. /Courtesy of Yonhap News Agency
Anthropic logo. /Courtesy of Yonhap News Agency

A study found that artificial intelligence (AI) could infiltrate smart home appliances such as TVs, refrigerators and robot vacuums on its own and use them as hacking footholds. It said AI models can run by combining the computing resources of multiple appliances even without data centers equipped with graphics processing units (GPUs).

Anthropic's red team researchers on the 12th released a report analyzing the possibility that an AI model could replicate itself and spread to Internet of Things (IoT) devices connected to the internet.

Home IoT devices such as TVs, refrigerators, ovens and robot vacuums have small memory and low computing performance, but they are relatively vulnerable to security threats and are constantly connected to the internet. The researchers said an AI model could operate by linking multiple devices into a single network and using each device's computing resources.

The researchers connected 15 Android operating system (OS)-based set-top boxes—sold on the market for about $55 (about 70,000 won) each—into a network, then distributed an AI model with 26 billion parameters, "Gemma4-26B-A4B," across them and succeeded in running it stably at a speed of 0.39 tokens per second.

Until now, cyberattacks on IoT devices have used malware without AI capabilities and had to be tailored to each device, but AI models can identify the characteristics of unfamiliar devices on their own, heightening hacking concerns. In addition, cyberattacks by AI using IoT devices are expected to be very difficult to block because there is no separate central point that can be controlled.

The researchers projected that by the end of this year, small AI models will have such capabilities.

The researchers likened this slow but steadily spreading AI to "digital slime mold" or a swarm of fruit flies. They warned that the emergence of these low-speed replicators could signal the arrival of faster and more aggressive AI replicators.

They emphasized that urgent countermeasures include abnormal traffic detection and quarantine by internet service providers (ISPs) and strengthened security by smart appliance manufacturers.

※ This article has been translated by AI. Share your feedback here.
Lee Ho-jun

Lee Ho-jun 편집국 테크부 기자

Lee Ho-jun 편집국 테크부 기자

Lee Ho-jun 편집국 테크부 기자

What to Read Next
What to Read Next
What to Read Next