Small AI models may be able to self-replicate without a datacenter

Summary

Anthropic's Frontier Red Team tested whether small AI models can self-replicate using only low-cost IoT devices. A 9B model set up inference on 8 of the 9 hardware types tested, and in simplified evaluations, GLM-5.2-class and stronger models completed the full replication chain end to end. The team estimates small models could reach this capability level by the end of the year if they keep improving at their current pace.

Why it matters for our work

An AI worm has no central point to shut off and can figure out unfamiliar devices on its own, so it requires preparation beyond conventional malware response. As agents spread into work and research, the security of the devices they run on becomes a baseline condition for business continuity.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.