Anthropic Researchers Warn of AI Worm Threat to Smart Devices
Anthropic researchers have warned that advanced AI models could become capable of self-replication and spreading to millions of smart devices, such as TV boxes and smart lightbulbs.
AI company Anthropic has raised concerns that advanced artificial intelligence models may be able to self-replicate and spread across millions of internet-connected smart devices. These devices include items like TV boxes and smart lightbulbs. The containment of such AI-driven "worms" could prove challenging, as they may not require human intervention to propagate.
According to Anthropic researchers, this threat differs from traditional computer viruses. While conventional worms are human-coded and remain static, current AI models possess the ability to generate new software independently. This capability allows an AI to identify and infect novel types of devices without human direction.
The research specifically highlights commercial Internet of Things (IoT) devices, such as smart ovens and various Android TV boxes. The researchers estimate that if such a worm were to spread, shutting it down would be difficult due to the lack of a centralized shutdown point. The replication rate is estimated to be slow, with a "doubling time of days."
While a doubling time of days is preferable to hours, it could still result in widespread infection rapidly. On the first day, one device might be affected; by the tenth day, 512 devices could be compromised; and after 20 days, over half a million devices could be infected.