Artificial intelligence models could be just months away from copying themselves across everyday internet-connected devices, including smart TVs and kitchen appliances, without human assistance, according to new research from Anthropic.
The company behind the Claude chatbot said smaller AI models are approaching the ability to find vulnerable devices, install copies of themselves, and keep spreading without relying on centralised data centres.
Get updated faster and for FREE: Download the Gulf News app now - simply click here.
The warning follows experiments examining whether AI systems could replicate across connected devices, including smart ovens and low-cost Android TV boxes.
Researchers found that models with fewer than 30 billion parameters could already complete many individual steps needed for self-replication, though they struggled to carry out the entire process reliably.
In simplified tests where researchers removed some security barriers, more capable models, including GLM-5.2, completed the full replication sequence.
Why smart home devices are vulnerable
Anthropic focused on internet-connected household and business devices because they often have weaker security protections than conventional computers.
The researchers found that a nine-billion-parameter AI model could successfully run on eight of the nine hardware types tested.
Devices such as smart televisions, ovens, refrigerators, and security cameras could provide computing power for smaller AI models, potentially letting them operate without access to expensive data centres.
Unlike conventional computer worms, AI-powered systems could also adapt to unfamiliar devices and work out how to run on different hardware.
Why self-replicating AI could be difficult to stop
Anthropic warned that AI models spreading across many connected devices would be hard to shut down because there would be no single central system to switch off.
Such systems might spread relatively slowly, potentially taking days to double in number, but could remain active across widely distributed devices.
Based on current rates of improvement, researchers estimate that smaller models could become capable of autonomous replication by the end of 2026, with models able to carry their own software and operating instructions potentially following soon afterwards.
However, Anthropic stressed that its experiments have not demonstrated a fully autonomous rogue AI system spreading across real-world household devices.
The company called for stronger security measures to prevent AI systems from exploiting vulnerable connected devices as their capabilities improve.