Tunis:Anthropic has issued a warning that small artificial intelligence (AI) models may soon have the ability to self-replicate on commercial Internet-of-Things (IoT) devices such as smart ovens and off-brand Android TV boxes, potentially achieving this capability by the end of the year.

According to Emirates News Agency, these self-replicating models pose a significant challenge due to the absence of a central switch-off point. Unlike traditional non-AI malware, these AI models could independently adapt to unfamiliar devices and potentially target data center processors.

Anthropic's experiments explored the potential for AI models to locate vulnerable devices, replicate themselves, and continue spreading autonomously. It was observed that small models with up to 30 billion parameters could perform many steps necessary for replication, though they struggled with completing the entire process independently.

In controlled experiments where some security measures were relaxed, the GLM-5.2 model and stronger variants managed to achieve end-to-end replication. Anthropic cautioned that if the improvement rate of small models continues as historically observed, they might reach this capability by the year's end, with even smaller weight-based replicators following closely.

The company emphasized the importance of enhancing system security to prevent the spread of autonomous AI agents. IoT devices, including Android TVs and smart appliances, were highlighted as particularly vulnerable due to typically weak security measures. It was also noted that such autonomous systems would likely replicate slowly, capitalizing on widely available but insecure computing resources.