Artificial intelligence company Anthropic has warned that increasingly capable small AI models could soon be able to replicate themselves across vulnerable internet-connected devices, raising concerns about a new class of cybersecurity threats involving everyday consumer electronics.
In a research report published on October 12, 2026, the company said its experiments showed that smaller AI models could perform many of the individual tasks required to identify vulnerable devices, install AI software and transfer model copies to other machines, although they often struggled to complete the entire process independently.
Also read: Brandy to headline Flytime’s House of R&B in Lagos
Anthropic cautioned that, if small models continue improving at their historical pace, they could reach the capability demonstrated in its simplified replication experiment by the end of 2026.
“Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things devices, like smart ovens and off-brand Android TV boxes,” the company said.
The research examined whether relatively inexpensive consumer devices could provide enough computing power to run AI models without relying on specialised data-centre infrastructure.
In one experiment, researchers ran a model identified in the report as Gemma4-26B-A4B across 15 low-cost Android TV boxes, each priced at approximately $55.
Working together, the devices achieved an inference speed of about 0.39 tokens per second, demonstrating how distributed computing resources could be combined to run a larger AI model.
Anthropic also tested models on tasks associated with potential self-replication, including locating devices, installing software, transferring model files and spreading copies to other machines.
Models with 30 billion parameters or fewer could complete many individual tasks but frequently struggled to coordinate the full sequence without assistance.
In another experiment, a nine-billion-parameter model successfully configured AI inference on eight of nine hardware types tested, indicating that smaller models could operate across a range of consumer devices.
The findings do not establish that autonomous AI self-replication is already widespread.
Rather, they highlight how improvements in models’ ability to carry out complex sequences of tasks could make such threats more plausible if attackers exploit vulnerable devices and networks.
Anthropic warned that a replicating system distributed across multiple devices could be difficult to contain because it would not rely on a single central system that could simply be switched off.
“A replicator like this would be challenging to shut down, since there’s no centralized shut-off point,” the company said.
Researchers estimated that, under the conditions examined, a replicator could take one to two days to spread to another host, using an inference-speed baseline of approximately 0.3 tokens per second.
The estimate reflects the experimental assumptions and should not be interpreted as a confirmed real-world infection rate.
To reduce the risks, Anthropic 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 called for stronger default security measures in consumer electronics to reduce the number of devices that could be exploited.
The warning comes amid wider concerns about the potential misuse of AI in cybersecurity.
The report referenced separate evaluations involving Anthropic and OpenAI models, in which researchers said AI systems had exploited security weaknesses under testing conditions.
In Nigeria, the Computer Emergency Response Team has also warned about cyberattacks involving botnets and vulnerable internet-connected devices.
Compromised servers, endpoints and Internet of Things devices can be recruited into coordinated attacks, making device security an increasingly important issue for households, businesses and network operators.
Anthropic’s findings add to the debate over how AI developers, device manufacturers, internet service providers and regulators should prepare for systems that can perform increasingly complex tasks with limited human intervention.
Also read: Brandy to headline Flytime’s House of R&B in Lagos
Although the experiments do not prove that small AI models can already autonomously replicate across consumer devices at scale, they underscore the importance of addressing security weaknesses before more capable systems make such operations easier to execute.
