Anthropic warns of AI ‘worms’ spreading across smart gadgets

Anthropic, an AI company, has warned that AI models may soon behave like computer worms, spreading between smart devices without human intervention. This finding highlights a new cybersecurity threat emerging from AI technology, which has already caused concern among governments and companies adopting AI rapidly. The research, published on Anthropic's blog, suggests that small AI models could soon be able to copy themselves from one internet-connected gadget to another, leveraging cheap, poorly secured consumer hardware such as smart TV boxes and kitchen appliances.

A botnet, a network of hijacked devices controlled by hackers, could serve as a potential target. The AI worm would likely be slow, taking days to copy itself, but it would be challenging to eliminate as there would be no central switch to turn off. While the estimate of how fast and far such AI malware could spread is based on limited data, the researchers at Anthropic believe that the amount of computing power in everyday devices and the size of AI models needed to carry out such attacks are expected to keep growing.

Experts caution that the biggest current threat is not AI systems going rogue but human bad actors using AI to enhance their hacking capabilities.