Anthropic Researchers Build Self-Spreading AI Worm on Smart TVs

Anthropic's internal offensive security team has demonstrated a working "AI worm" — a self-replicating malicious program powered by artificial intelligence — that spread on its own across a network of 15 internet-connected smart televisions, the company disclosed this week. The finding marks a shift from theoretical AI danger scenarios toward a concrete, documented vulnerability that security professionals say deserves immediate attention.

Unlike conventional malware, AI worms can leverage machine-learning capabilities to adapt and propagate without direct human instruction. The Anthropic red team's program overcame a significant technical barrier: rather than requiring a single, high-powered computing device to run the AI model, the worm distributed its workload across multiple lower-powered televisions, drawing a small amount of processing capacity from each to remain functional as a whole.

Read more Onsemi in Talks to Acquire Chip Startup Ferric for $1 Billion →

The demonstration matters because it suggests that resource constraints — long considered a natural brake on AI-powered malware spreading through consumer electronics — may be less of an obstacle than previously assumed. Anthropic researchers noted that while today's edge devices still lack sufficient computing power to make such attacks trivially easy, current hardware trends could make the approach broadly feasible by the end of this year.

The disclosure arrives at a moment when the volume of AI-related security warnings has outpaced many experts' ability to assess relative risk. By grounding their findings in a working proof-of-concept rather than projections, Anthropic's red team aimed to focus industry attention on a threat that requires no human-like reasoning or catastrophic capability — only incremental improvements to technology already on a predictable development curve.

Continue reading at Business | The Guardian.