Anthropic Researchers Build AI Worm That Spreads Across Smart TVs
Anthropic's red team created a self-replicating AI program that spread across 15 internet-connected televisions, flagging a near-term security risk.
Anthropic's internal offensive security team has developed a self-replicating artificial intelligence program — commonly called an "AI worm" — that successfully propagated across internet-connected smart televisions without human intervention, the company disclosed this week. The demonstration marks a concrete, technical milestone in a threat category that has until now remained largely theoretical.
The research team found a way around a key obstacle: AI programs typically require significant computing power to function. Their solution was to distribute the workload across 15 televisions simultaneously, drawing modest processing capacity from each device to keep the malicious program operational. That architectural approach allowed the worm to sustain itself on hardware far less powerful than conventional AI infrastructure.
Read more Waitrose Opens First New Supermarket in Eight Years in London →
Anthropic noted that for AI worms to succeed broadly, models need to become capable enough to run efficiently on lower-powered consumer devices. The company warned that current trends in AI miniaturization make that threshold realistically achievable before the end of 2025, potentially opening a wide attack surface across the billions of ordinary connected devices in homes and offices worldwide.
The disclosure arrives against a backdrop of accelerating AI security warnings that even specialists struggle to prioritize. What distinguished Anthropic's report, observers noted, was its grounding in a working proof of concept rather than speculative risk modeling — a practical demonstration that autonomous, device-hopping AI malware is no longer a distant concern but an emerging engineering reality.
Continue reading at Business | The Guardian.
