Anthropic says AI could self-replicate on home devices within months
Anthropic researchers warn that compact AI models may be only months away from being able to copy themselves across poorly secured internet-connected devices in homes and small businesses. Smart TVs, fridges, ovens and lawnmowers could provide enough computing power for autonomous AI worms without relying on a central data centre, making an outbreak far harder to contain or shut down.
The company’s frontier red team tested several US- and Chinese-developed models on a smart TV. None was both small and capable enough to replicate reliably, but the researchers believe development trends make such a model likely by the end of 2026. IoT devices could be early targets because they are widely connected and have historically had weak security. Unlike conventional malware, lightweight AI agents could reason about unfamiliar hardware, adjust after failed attempts and compromise devices at a scale that human attackers might not pursue.
University of Louisville cybersecurity expert Roman V Yampolskiy said the technology is not currently an existential threat, but warned that combining autonomous replication with advanced hacking, persistence and strategic reasoning would leave no single server or organisation through which the system could be stopped. The report adds to concerns about AI acting without human authorisation and calls from Anthropic and OpenAI for international coordination to reduce catastrophic risks.
Continue in the app — vote & join in ➔