WASHINGTON: Artificial intelligence (AI) research company Anthropic has warned that smaller AI models could be just months away from autonomously replicating themselves across networks of commercial Internet of Things (IoT) devices, Reuters reported.
In a series of simulated experiments, Anthropic’s safety team examined whether malicious AI agents could identify vulnerable devices, copy AI models onto the hardware and spread across connected networks without human intervention.
The research found that AI models with fewer than 30 billion parameters (30B) could already perform many of the technical steps required for self-replication.
However, Anthropic said the models still faced challenges completing the entire replication process consistently in real-world environments, where systems are more complex and security protections remain in place.
In simplified simulations with several security safeguards removed, models such as GLM-5.2 and other more capable AI systems were able to complete the replication process.
Anthropic warned that self-replicating AI could pose a difficult cybersecurity challenge because it might spread without a central point of control and adapt to unfamiliar hardware, unlike conventional malware.
The study focused on consumer IoT devices, including smart ovens and unbranded Android TV boxes, which may lack robust cybersecurity protections.
The company cautioned that if smaller AI models continue improving at their current pace, they could potentially achieve autonomous self-replication by the end of 2026. It also warned that replication based on model weights could emerge in the near future.
According to Anthropic, networks of autonomous AI agents could spread gradually by exploiting widely available computing resources across the internet.
The findings underscore the need for the technology industry to strengthen cybersecurity measures across smart-device ecosystems before such capabilities become more advanced.
-AWANI
