New AI Models Are Replicating Themselves

According to research by an AI company, small models will soon be able to replicate themselves on internet-of-things devices without access to data centers.

According to a research conducted by an artificial intelligence company, small models will soon be able to reproduce themselves on IoT-based devices without access to data centers.

According to Ambassador Arjan, quoted by Reuters, Anthropic announced that its tests show that small artificial intelligence models may be just months away from achieving the ability to replicate themselves on commercial IoT devices, such as smart ovens and various Android TV boxes.

In its experiments, Anthropic investigated whether AI models could find vulnerable devices, copy themselves to these devices, and continue to spread without human help. The results showed that small models with less than 30 billion parameters can go through many steps required for reproduction.

Of course, replicating models were challenged to complete the sequence from the starting point to the end of the replication. Anthropic found that GLM 5.2 AI and more powerful models were able to reproduce head-to-head in its simplified tests where some security steps were removed. The company also warned that replicating models are difficult to shut down because, unlike non-artificial intelligence malware, there is no central shutdown point, and said models can identify unknown devices on their own and target data center processors.

In its tests, the company focused on Internet of Things devices such as Android TVs, smart devices, and other Internet-connected gadgets because these devices usually do not have proper security.

Entropic also warned that if small models are upgraded at their historical rate, they could be self-replicating by the end of this year, and replicators based on light weight models will follow suit.

Such AI models are likely to be slow to replicate and rely on widely available and somewhat insecure resources.