SAN FRANCISCO, US – Artificial intelligence models may be only months away from developing the ability to spread like computer worms, copying themselves between internet-connected devices without human intervention, Anthropic has warned.
The potential development could open a new front in cybersecurity threats linked to artificial intelligence, raising concerns as governments and technology companies accelerate the adoption of AI-powered systems.
In research published on its blog late Monday, Anthropic examined how smaller AI models could potentially spread across consumer electronics, including smart TVs, phones, laptops, cars and household appliances.
Unlike conventional AI systems that rely on powerful data centres, these models could operate across networks of inexpensive, poorly secured devices, such as off-brand Android TV boxes and connected kitchen appliances.
A computer worm is malicious software capable of spreading independently from one machine to another. An AI-powered version could potentially use the computing resources of compromised consumer devices to continue propagating without depending on a central data centre.
Anthropic Examines Lessons From Kimwolf Botnet
The researchers based their estimates partly on Kimwolf, a botnet that infected more than two million devices at its peak, primarily off-brand Android TV boxes.
A botnet is a network of compromised devices controlled by attackers, often without the owners’ knowledge.
Anthropic researchers William Zhou and Daniel Freeman said an AI worm of the type they examined would probably spread relatively slowly, potentially taking days to replicate across devices.
They compared the process to the gradual spread of slime mould or a swarm of fruit flies, rather than the rapid propagation often associated with conventional computer worms.
However, the researchers warned that eliminating such a network could prove difficult once it became established because it might not depend on a single central system that could be switched off.
Researchers Caution Against Treating Estimate As Certainty
Anthropic acknowledged that its estimate was preliminary, noting that the analysis focused on one type of consumer device and one historical botnet.
The research therefore does not establish that an AI worm capable of spreading autonomously across consumer devices already exists or will necessarily emerge within the projected timeframe.
Nevertheless, the researchers said the underlying conditions could become more favourable over time. The computing power available in everyday consumer gadgets is expected to increase, while AI models capable of performing increasingly complex tasks may become smaller and more efficient.
As technology companies move to embed AI directly into consumer electronics, the findings highlight the importance of securing connected devices against potential forms of automated cyberattack.
