After calling out the tech industry to slow down the pace of AI development, Anthropic has come up with another warning. The tech giant said in a recent blog post that small AI models may soon be able to replicate themselves across internet-connected devices. This raises concerns about a new type of cyberattack that may become difficult to contain.
According to Anthropic, its experiments suggest that small models may acquire the ability to copy each other on everyday devices like smart ovens and low-cost Android TV boxes, within months.
How Small AI Models May Spread Across Devices
The research from Anthropic focuses on autonomous replication and adaptation (ARA), a process in which AI models spread on their own without any human intervention. The company explained, “Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things (IoT) devices, like smart ovens and off-brand Android TV boxes.”
Anthropic revealed that its tests found a model with nine billion parameters may get an AI inference system running on eight of the nine hardware types tested. The researchers have also demonstrated that a model with 26 billion parameters may run across 15 inexpensive Android TV boxes, each costing nearly $55. The setup delivered approximately 0.39 tokens per second.
Low-cost IoT devices often have limited memory and weak security protections. Anthropic noted that multiple devices could work together, with each running a portion of a larger AI model.
It explained, “Only GLM-5.2 out of our set of tested open-weight models is capable of full end-to-end replication (assuming its capabilities could be packed into Llama3.1-8B). Every subsequent infection is the worm itself hopping node to node over an unauthenticated ADB port, before setting up an inference server to serve Llama3.1-8B. We don’t yet have a full rogue replicator, but agents are becoming increasingly capable of completing the individual steps in the replication chain (and passing on their adaptations to future copies of themselves).”
The company also warned that a large network of infected devices could spread slowly over several days while remaining difficult to eliminate completely. To reduce the risk, Anthropic recommended stronger default security settings, improved network monitoring and isolating infected devices. It stressed that AI-powered self-replicating malware has not yet been demonstrated as a complete, unrestricted real-world threat.
