Smart homes, especially those from lesser-known brands, have for some time been a potential vector for hackers to launch web-based cyberattacks or gain unwanted network access.
But artificial intelligence adds a new dimension to the problem. Malicious AI models could soon find their way into Internet of Things devices, from smart stoves and TVs to smart refrigerators, and self-replicate across other gadgets, according to a study by Anthropic (ANTH.PVT).
Such models would be harder to shut down because there is no central hub to take them offline. And while they can move slowly from device to device, the company says the models can figure out how to spread across devices they’re unfamiliar with.
The announcement comes after Apple (AAPL) launched its HomeView, putting the smart home market back in focus.
Anthropic conducted its testing using Android-based devices.
It also follows a report that hackers broke into internet-connected refrigerators in French grocery stores, raising temperatures, ruining food and starting fires.
During its own testing, Anthropic said it found that larger AI models can split and run across multiple devices. Some models have also managed to transmit knowledge to replicated versions of themselves, allowing them to adapt to different hardware.
“Machines that could run a large model fragment for a <30B model include smart refrigerators, smart ovens, automatic vacuum cleaners, and lawn mowers,” the company said.
Additionally, Anthropic notes, new smart devices incorporate AI chips to power their own AI models, giving attack models more power to continue spreading.
And thanks to their ability to run on different types of devices, they could eventually spread to increasingly powerful machines.
Anthropic says that malicious replicators are not a problem at the moment, since models with fewer than 30 billion parameters can perform individual steps in the replication process but make errors when executing the entire chain of actions.
Nevertheless, the company claims that at the current rate of improvement of the model, it may be able to achieve such capabilities by the end of 2026.
Email Daniel Howley at dhowley@yahoofinance.com. Follow him on Twitter at @DanielHowley.
Click here for the latest technology news that will impact the stock market
Read the latest financial and business news from Yahoo Finance
Gn world
