Anthropic said on Monday that its experimental findings suggest small AI models may be only months away from gaining the ability to self-replicate on commercial Internet of Things (IoT) devices such as smart ovens and lesser-known Android TV boxes.

In the experiments, Anthropic examined whether AI models could identify vulnerable devices, copy themselves onto those devices, and continue spreading without human intervention.

Anthropic found that small models with fewer than 30 billion parameters were able to complete multiple steps required for replication.

However, the company also found that such AI models struggled to carry out the entire operational process from beginning to end.

In simplified tests that removed certain safety guardrail steps, Anthropic found that GLM-5.2 and more powerful models were able to achieve end-to-end replication.

Anthropic warned that models with self-replication capabilities would be difficult to shut down because they have no centralized shutdown control point; in addition, unlike non-AI malware, such models can also identify unfamiliar devices on their own and target data center processors as attack targets.

The company's research focused on IoT products such as Android TVs, smart appliances, and other connected devices, because these devices typically lack sufficient security protection.

The company also warned that if small models continue improving at their historical pace, they will be able to achieve autonomous replication before the end of this year, with weight-based small replicators also emerging shortly afterward.

Anthropic also noted that the replication process of such autonomous AI systems may be relatively slow and dependent on widely available but comparatively less secure computing resources.

The company added that system defenses should be strengthened to make it more difficult for such AI agents to spread.