Quick Takeaways
- AI-powered worms are a future risk, not a practical threat today.
- Anthropic showed cheap Android TV boxes can host a large AI model, and tested agents’ ability to spread across a network.
- Only the strongest model reliably reached most devices—and the test used unusually easy, unauthenticated access.
- Model growth could raise the risk, but bandwidth, storage, and device constraints remain major hurdles; defenders should prepare for automation.
Anthropic Finds Ingredients for Future AI Worms
Researchers Test How an AI Worm Could Spread
Anthropic researchers tested whether AI models could complete steps that a self-spreading computer worm would need. Those steps included finding connected devices, moving software to them, and continuing the process.
In one experiment, researchers ran a large AI model across 15 inexpensive Android TV boxes. The test showed that low-cost devices could work together to host an AI model, although the system ran slowly.
In a separate test, AI models tried to spread a software payload across an isolated network of eight Android devices. One model reached all eight devices in four of five trials. However, the researchers gave the models unusually easy access. The test therefore showed a possible capability, not a realistic attack.
Real-World Barriers Remain
The findings suggest that AI systems can already handle many tasks a worm might require. Still, no test showed that a low-cost model could copy itself and spread independently across real-world networks.
Practical obstacles remain. For example, devices may lack enough storage, and slow connections could make it difficult to transfer the large files that AI models need. Security controls also make unauthorized access harder than it was in the experiment.
Even so, the research offers an early warning. As AI tools improve, defenders may need faster ways to spot and contain automated attacks. That work can help society gain the benefits of more capable technology while limiting risks to the connected devices people rely on every day.
Discover More Technology Insights
Learn how the Internet of Things (IoT) is transforming everyday life.
Access comprehensive resources on technology by visiting Wikipedia.
CyberRisk-V1
