Artificial intelligence models may be just months away from gaining the ability to copy themselves across everyday internet-connected devices, including smart ovens, Android TV boxes and other household gadgets, without human intervention, AI company Anthropic warned Tuesday. Such technology could create a new type of self-replicating digital threat that would be particularly difficult to eliminate because it would have no central system that could simply be switched off.
The warning follows experiments conducted by Anthropic researchers to determine whether AI systems could autonomously identify vulnerable devices, install copies of themselves and continue spreading to other machines. The research focused on smaller AI models capable of operating on relatively inexpensive hardware rather than relying on the powerful processors typically found in data centers.
Self-replicating AI could spread through smart TVs and ovens within months, Anthropic warns
(Photo: Reuters)
According to Anthropic, models containing 30 billion parameters or fewer, a measure of their size and complexity, can already complete many of the individual tasks required for autonomous replication. However, the researchers found that these smaller systems still struggled to combine all the necessary steps into a reliable, uninterrupted process, frequently making mistakes before completing the full sequence.
In simplified experiments that removed some security obstacles, more powerful models, including GLM-5.2, successfully completed the entire replication process from beginning to end. The findings suggest that the main barrier is no longer whether AI can perform the individual tasks, but whether sufficiently small models can coordinate them reliably without human assistance.
Why smart home devices could become the next target
Anthropic focused on internet-of-things (IoT) devices, a category that includes smart televisions, connected kitchen appliances and other electronic products linked to the internet, because many have relatively weak security protections. Unlike sophisticated data center infrastructure, these devices are often inexpensive, widely distributed and less frequently updated or monitored for potential intrusions.
The concern is that an autonomous AI system capable of exploiting vulnerable devices could use their computing resources to run copies of itself and identify additional machines to target. Unlike conventional malware, which generally follows predetermined instructions, an AI-powered system could potentially adapt to unfamiliar hardware and operating environments, allowing it to spread across different types of devices without requiring specific instructions for each one.
Such a system could also prove far more difficult to shut down. Because copies would operate independently across numerous devices, disabling one infected machine would not necessarily stop the others. Anthropic warned that these capabilities could eventually extend beyond consumer electronics to more powerful computing systems, including processors used in data centers.
A threat that could emerge before the end of 2026
Anthropic said that if smaller AI models continue improving at their historical rate, they could reach the level of capability needed for autonomous replication by the end of 2026, with compact models capable of carrying their own computational components to new devices potentially following soon afterward.
The company emphasized, however, that the predicted threat remains based on experimental findings rather than a demonstrated outbreak involving autonomous AI systems spreading across the public internet. It also noted that any early self-replicating systems would likely spread relatively slowly, relying on the limited computing power available from widely distributed, poorly protected devices.
Nevertheless, the researchers warned that the potential emergence of such systems presents a growing cybersecurity challenge. Anthropic urged stronger security protections for connected devices and computing infrastructure to make it harder for autonomous AI agents to establish themselves, reproduce and spread beyond human control.