Self-replicating AI: Is your fridge a cybersecurity risk?

In modern households, connected electronic devices, known as Internet of Things (IoT), are increasingly common. A recent report from Anthropic, an AI company, warns that AI-powered malware could infiltrate these devices, replicate themselves, and cause harm. AI worms, unlike traditional computer worms, are more complex and adaptable, capable of solving problems, coping with different operating systems, and potentially learning how to bypass security updates.

Anthropic's researchers demonstrated this concept by distributing an AI model across 15 low-cost Android TV boxes. The model was able to infiltrate one of the simulated devices, replicate itself, and infect all eight additional devices in four out of five attempts. If equipped with malware, such an AI model could potentially adapt quickly to new environments, disrupt processes, and learn to circumvent security measures.

The potential for an AI worm to create a large botnet, with one infected device becoming a source of many compromised devices, is particularly concerning. The malware could continuously learn as it spreads, potentially jumping to different types of electronic devices running different software. Even if infected devices are disconnected from the network, the malware's decentralized structure would make it difficult to contain.

In a real-world example, attackers in France tampered with refrigeration and freezer systems at numerous supermarkets, causing food to spoil and fires. The attack exploited outdated security software and may have used an AI worm. While it is not confirmed whether a self-replicating AI worm was used, Anthropic's research shows the significant potential for AI to facilitate such attacks.

The French cybersecurity agency advised companies to regularly update their connected devices, install essential security patches, and disable unnecessary internet access.