Anthropic red team: small AI models may self-replicate on IoT devices within months

ns123abc · x · 2026-10-13

Anthropic's Frontier Red Team published an evaluation of autonomous replication and adaptation (ARA) in AI models.

Key finding: small AI models may be months away from self-replicating on commercial IoT devices like smart ovens and off-brand Android TV boxes. Such a replicator has no centralized shutdown point, though it would be slow (doubling time of days); unlike classic malware, it could figure out unfamiliar devices on its own.
Eval results: a 9B model got inference running on 8 of 9 hardware types tested. Models of 30B and under can already complete many individual replication steps but make mistakes composing them end to end. In simplified evals, GLM-5.2 and stronger models can replicate end to end; at historical improvement rates, small models will match this capability by the end of the year, with small weight-based replicators possible soon after.
Threat shape: an "AI worm" could carry model weights (or re-download them) and run inference on compromised hardware — the weights become part of the malware payload, with datacenter GPUs as high-value targets.

A concrete empirical assessment of frontier-model loss-of-control risk, not mere speculation.

More from Models