Anthropic warns: “AI could self-replicate through smart appliances” [Global AI Briefing]

[Financial News] A warning has been issued that a cyber threat could become reality in which artificial intelligence (AI) infiltrates internet-connected appliances such as smart TVs and refrigerators, replicates itself, and spreads to other devices.
The concern is that AI could evolve into an “AI worm” that replicates and spreads on its own, causing new cyberattacks. This makes security checks urgent not only for AI but also for smart appliances and network equipment.
According to a report by The Guardian on the 13th (local time), Anthropic’s Frontier Red Team warned in an official research report that “small AI models can run outside data centers, on ordinary Internet of Things (IoT) devices,” and that “AI could acquire the technical ability to spread itself within the year.” It added that smart TVs, smart refrigerators, ovens, and security cameras could become potential targets of AI worm attacks.

Unlike conventional computer viruses, an AI worm can use an AI model to replicate itself without being instructed to do so and run on other devices. It infiltrates an internet-connected device with weak security, secures an environment in which to run, then finds other devices and repeats the same process.
Anthropic’s Red Team said it successfully ran AI models on real smart appliances in an experiment, distributing them across 15 Android TV devices with 4GB of memory each.
The researchers explained that “small AI models can perform most of the individual tasks needed for replication, and in simplified experiments, some models also showed the ability to link the replication process together from beginning to end.”
The researchers also said that “an autonomous self-replicating AI capable of flawlessly completing every step, from exploiting security vulnerabilities in real devices to replicating and running the model, has not yet been developed.” They nevertheless projected that “if the capabilities of small AI models continue to improve at their current pace, they could reach the relevant level within the year.”
Security experts say that if AI worms become a reality, it would be virtually impossible to determine the full scope of infected devices and block an attack all at once. They say it would be difficult to stop an AI worm from replicating itself or carrying out attacks simply by cutting power to a single data center, and that it would also be difficult to identify every low-cost device that has not received proper security updates.
They noted that it could also become harder to detect and respond to AI using existing security systems, particularly if it adapts how it runs on each device.
With a recent AI-enabled personal data breach in the financial sector throwing the country into turmoil, experts are increasingly urging urgent security checks on internet-connected hardware in South Korea, including smart appliances and outdated network equipment, to prepare for the emergence of AI worms.

[email protected] Lee Gu-soon Reporter