Anthropic said on Monday that its experiments suggest small AI models may need only months before they can self-replicate on commercial Internet of Things devices such as smart ovens and obscure-brand Android TV boxes. According to Sina Finance, the company said it tested whether AI models could find vulnerable devices, copy themselves onto them, and keep spreading without human intervention.

Anthropic found that small models with fewer than 30 billion parameters could complete several of the steps needed for replication, but they still struggled to carry out the full process from start to finish. In simplified tests that removed some safety safeguards, Anthropic said GLM-5.2 and more powerful models were able to replicate end to end.

Anthropic warned that self-replicating models would be difficult to shut down because they lack a centralized kill switch. It also said that, unlike non-AI malware, such models can identify unfamiliar devices on their own and target data center processors. The company said its research focused on IoT products such as Android TVs, smart appliances, and other connected devices because they often lack sufficient security protections.

Anthropic also warned that if small models continue improving at their historical pace, they could achieve autonomous replication by the end of this year, with weight-based small replicators following soon after. The company added that such autonomous AI systems may replicate slowly and rely on widely available but relatively insecure computing resources, and it said system defenses should be strengthened to make their spread harder.