Researchers at Anthropic have tested how small AI models might help software spread across unfamiliar connected devices. Their findings raise questions about smart TVs and other internet-connected gadgets—but they have not demonstrated an AI worm autonomously spreading through real homes.
Your smart TV, streaming box or home camera may not look like a computer worth attacking. But many connected devices run software, connect to the internet and receive security updates less frequently than laptops and phones. Researchers are now examining whether AI could make it easier for malicious software to adapt to this uneven landscape.
In research published on October 12, Anthropic’s Frontier Red Team explored whether an AI model could help create what the company calls a self-replicating AI system: software that carries a model and the components needed to run it, then attempts to establish copies on other devices.
What did the researchers actually test?
Anthropic said a model with nine billion parameters was able to get an inference server running on eight of nine hardware types in its evaluation. The researchers also found that some stronger models could complete a sequence of replication steps in simplified tests.
The experiments included inexpensive Android TV boxes and other constrained hardware. The idea was to test whether a small model could be packaged with the software needed to run it, then configured on a different device—rather than relying on a large, central data centre for every step.
These results are meaningful, but they need context. The research does not establish that an AI worm is currently roaming the internet and infecting smart TVs. Anthropic said the full process remains difficult to complete reliably: smaller models make mistakes when combining the steps, while more capable models can be too demanding for inexpensive devices.
Why would a smart TV be vulnerable?
Smart TVs and streaming boxes are part of the Internet of Things (IoT), the broad category of devices that connect to networks to provide services. Many are based on familiar operating systems, and some expose debugging tools or network services that should not be accessible to strangers.
Security risks can grow when devices are no longer receiving updates, use weak or reused passwords, run outdated software, or are placed on the same network as work computers and personal files. An attacker does not necessarily need a sophisticated AI system if a device already has a known, unpatched vulnerability.
AI could change the economics of an attack by helping software investigate unfamiliar environments and adapt its next steps. That is the concern researchers are exploring—not a claim that AI has made every smart TV easy to compromise.
Is an AI worm different from ordinary malware?
Traditional worms spread by following programmed instructions and exploiting specific weaknesses. An AI-assisted worm could, in principle, use a model to interpret a new environment, choose among tools and adapt when a familiar method fails.
That flexibility could make some attacks harder to predict. But AI also brings practical limits: models need computing resources, replication requires network access and storage, and an unreliable model may fail before it can spread. Anthropic’s study describes a developing capability, not a finished consumer-device threat.
What can you do to protect your connected devices?
- Install firmware updates. Check the TV, streaming box, router and camera for updates from the manufacturer.
- Replace devices that no longer receive security support. An old device that cannot be patched can remain a weak point on the home network.
- Change default passwords. Use unique credentials and enable multi-factor authentication wherever the device supports it.
- Turn off features you do not use. Disable remote administration and debugging services unless they are genuinely needed.
- Secure the router. Update its firmware and consider a guest or IoT network for smart devices, separate from computers that hold sensitive information.
- Buy with support in mind. Before purchasing a connected device, check how long the manufacturer promises software and security updates.
Should you be worried right now?
There is no reason to assume your television has been infected with self-replicating AI malware. The research is better understood as an early warning about what might become possible as models improve and connected devices remain poorly secured.
The practical lesson is familiar: update devices, remove unnecessary access and keep less-trusted gadgets away from sensitive systems. AI may help attackers adapt more quickly, but it does not remove the need for an actual route into a device.
For now, the most important finding is not that AI malware has taken over the smart home. It is that researchers are testing the building blocks of a more adaptable threat—and that basic device security will matter if those building blocks become more reliable.
