Anthropic Researchers Build Self-Spreading AI Worm on Smart TVs
Anthropic's red team created a malicious AI program that spread autonomously across 15 smart TVs, raising alarms about a new class of practical cyber threat.
Anthropic's internal offensive security team has demonstrated a working "AI worm" — a self-replicating malicious program powered by artificial intelligence — that spread autonomously across a network of internet-connected smart televisions, the company disclosed this week. The research marks one of the first documented cases of such a program operating on consumer-grade devices rather than high-powered servers.
The key technical hurdle for AI worms has historically been computational demand: running AI models typically requires significant processing power. Anthropic's red team circumvented that barrier by distributing the program across 15 smart TVs simultaneously, drawing small amounts of processing capacity from each device to keep the worm functional. The approach suggests that a distributed architecture could allow AI-driven malware to operate well below the hardware thresholds previously considered necessary.
Read more Waitrose Opens First New Supermarket in Eight Years in London →
The company warned that the convergence of two trends — AI models becoming more capable and edge devices becoming more powerful — could make such attacks broadly feasible before the end of the year. That timeline lends urgency to findings that might otherwise be dismissed as theoretical. Security professionals are already struggling to prioritize a flood of AI-related threat disclosures, but Anthropic's report stands out for its grounded, technical framing, deliberately stripped of the speculative or anthropomorphic language common in the field.
The worm's demonstrated ability to propagate without human direction across ordinary household hardware signals a potential shift in the cyberattack landscape. Unlike conventional malware, an AI-driven worm could adapt its behavior in response to defenses it encounters, complicating detection and containment. Researchers say the smart-television proof of concept is a starting point, and similar techniques could in principle be applied to other connected devices in homes and offices.
Continue reading at Business | The Guardian.
