Internet-connected appliances and other devices that exchange data through the Internet of Things could face a new cybersecurity threat as artificial intelligence advances. According to a report published by Anthropic in October, AI-powered malicious software could install itself on these devices and spread autonomously.
Traditional computer worms move from one machine to another by exploiting known vulnerabilities or weak access controls. AI-based worms could be more difficult to contain because they might solve problems independently and adapt their behavior to different types of devices.
Anthropic researchers demonstrated the concept in a controlled test involving simulations of eight Android devices. The tested AI model installed itself on one device, replicated from it and infected the others. It reached all eight targets in four of five attempts and infected seven in the remaining test.
Anthropic developers said that, if combined with malicious software, a worm using this type of AI could learn to bypass security measures in approximately one to two days. The company also said that such software would not necessarily require a large data center to operate.
The infection could potentially turn compromised devices into a botnet. Although an individual connected device has limited processing power, a growing number of infected devices could allow the malware to spread faster. The decentralized structure of such a network would make containment more difficult, even if devices were disconnected and their security flaws were fixed.
Anthropic compared the potential spread to slowly growing digital slime mold. The company believes smaller AI models could reach this level of dissemination capability during 2026.
Sources: Olhar Digital.