Anthropic Warns: AI Models May Soon Begin Self-Replication

Anthropic stated that its experiments indicate that small AI models could, within a few months, become capable of self-replicating on commercial Internet of Things (IoT) devices—such as smart ovens and Android-powered televisions from lesser-known brands.

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

According to Reuters, the company found that small models—those with fewer than 30 billion parameters—could complete the various steps required for self-replication.

However, the company noted that these AI models struggled to complete the entire sequence from start to finish.
In simplified tests where certain security measures were removed, Anthropic found that the GLM-5.2 model—and more powerful models—were capable of fully self-replicating from beginning to end.

Anthropic warned that self-replicating models would be difficult to stop, as there is no central control point to disable them; furthermore, unlike traditional malware, they are capable of independently discovering unfamiliar devices and targeting data center processors. The company focused its research on Internet of Things (IoT) devices—such as Android-based televisions, smart home appliances, and other internet-connected devices—noting that these devices often lack adequate security.

The leading AI company also warned that if small models continue to improve at their current unprecedented pace, they will be capable of autonomously replicating themselves by the end of this year, with weight-based replication models likely to follow suit shortly thereafter.

Anthropic further noted that these autonomous AI systems would likely replicate slowly, relying on widely available computing resources that often lack sufficient security. The company added that systems should be strengthened to make the proliferation of such AI agents more difficult.