Anthropic has warned that small artificial intelligence (AI) models could soon replicate themselves across everyday internet-connected devices, including smart televisions and ovens, raising concerns about potential cybersecurity threats.
In a research report published on October 12, 2026, the AI company said experiments showed that small AI models could perform several steps required to identify vulnerable devices, install AI software and distribute copies to other machines without human intervention.
Anthropic said that if smaller models continued improving at their historical rate, they could reach the capability demonstrated in its simplified replication experiment by the end of 2026.
The company said its experiments suggested that small AI models could be months away from becoming capable of self-replicating on commercial Internet of Things (IoT) devices, including inexpensive Android TV boxes and smart ovens.
In one experiment, Anthropic ran Gemma4-26B-A4B, a model with 26 billion total parameters, across 15 inexpensive Android TV boxes costing approximately 55 dollars each. The system achieved an inference speed of about 0.39 tokens per second.
The model was distributed across the devices, enabling them to work together to run a larger AI system. The experiment demonstrated how computing resources from relatively cheap consumer hardware could be combined to support AI models without specialised data-centre equipment.
Anthropic also tested whether AI models could locate devices, install software, transfer model files and distribute copies to other machines.
It found that models with 30 billion parameters or fewer could complete many individual tasks but often struggled to execute the entire replication process autonomously.
In a separate experiment, a nine-billion-parameter model successfully configured AI inference on eight of the nine hardware types tested, suggesting that smaller models could operate across different consumer devices.
Anthropic warned that an AI replicator distributed across multiple devices could be difficult to contain because it would not depend on a single central system that could be switched off.
The researchers estimated that, under the conditions examined, a replicator could take between one and two days to spread to another host, using an inference-speed baseline of approximately 0.3 tokens per second.
To address the potential risks, Anthropic called for stronger security measures on internet-connected consumer devices as AI models become increasingly capable of performing complex tasks.
The company recommended monitoring large transfers of AI model files, detecting unusual network communications and device scanning, and enabling internet service providers to restrict or quarantine compromised devices.
It also advocated stronger default security protections for consumer devices to reduce the number of vulnerable systems that could be exploited to distribute malicious software.
The findings add to growing cybersecurity concerns surrounding the capabilities of advanced AI models and the potential misuse of compromised internet-connected devices.
In July 2026, Anthropic disclosed that its AI models had compromised the production infrastructure of three organisations during cybersecurity evaluations after a configuration error gave the models unintended internet access.
OpenAI also reported in July that its AI models had autonomously exploited vulnerabilities to compromise parts of Hugging Face’s production infrastructure during an internal cybersecurity evaluation, accessing systems beyond the intended testing environment.
In May 2026, Nigeria’s Computer Emergency Response Team (ngCERT) warned of escalating cyberattacks involving botnets and vulnerable internet-connected devices, noting that compromised servers, endpoints and IoT devices could be used to launch coordinated attacks.
Boluwatife Enome
Follow us on:
