Truace
Subscribe

What are self-replicating ‘AI worms’, and why is Anthropic worried about them?

Warnings about what artificial intelligence might do in the future – together with alarming reports of what it has gotten up to already – are being issued so quickly that even security experts have difficulty triaging what to worry about. Yet a report this week from Anthropic’s offensive research team grabbed attention by describing a new, practical threat, one devoid of human characteristics or apocalyptic musings. The red team, as such internal testers are called, said it created a malicious AI program that sp…

By the Truvace desk2 min read1 sourceJournalismTRV-2026-1825· cite & verify
What are self-replicating ‘AI worms’, and why is Anthropic worried about them?

In brief

Anthropic's offensive research team reported building a malicious AI program that moved by itself through smart televisions. It split parts of itself across 15 sets and ran by drawing a small amount of power from each, getting around the usual need for heavy computing capacity.

The case matters because internet-of-things devices tend to be patched rarely, and past worms exploited unpatched flaws to cause enormous damage. Its real-world danger is disputed: some experts call it stunt hacking, the demonstration is slow, and basic security hygiene is said to stop most worms.

Main points

  1. The red team's AI worm spread autonomously across smart televisions and could do the same in other ordinary devices.
  2. Researchers theorized an AI worm could adapt to new hardware by researching newly disclosed vulnerabilities.
  3. Anthropic said a successful worm would need more capability on weaker devices than is possible now, but trends could make that feasible by year's end.
  4. Experts such as Guan and Guerrero-Saade argued the threat is slow, detectable, and overhyped, while Saxe and Herbert-Voss urged preparation.

The problem

Anthropic's red team built an AI worm that spread autonomously across smart televisions, showing how poorly patched connected devices could become hard-to-stop vectors for self-replicating attacks.

The rundown

The article places the finding in the history of worms, citing the Morris worm of 1988, Stuxnet, NotPetya and WannaCry, several of which relied on known vulnerabilities that owners had not patched. NotPetya and WannaCry are said to have caused tens of billions of dollars in damage.

Experts differ on urgency. Joshua Saxe said the cybersecurity community is not taking such risks seriously enough, and Ariel Herbert-Voss said existing infrastructure is not ready. Chris Inglis suggested worms could also be used for benevolent purposes and said the question is how they will be employed and countered.

A University of Toronto paper reportedly infected a full-size computer and spread to similar machines faster than the television worm. Jonas Guan said capabilities are developing rapidly but that the Anthropic version moves very slowly and can be blocked at multiple layers. The article also notes skepticism that AI company warnings serve marketing or policy aims, and that US oversight efforts remain fragmented.

Sources

  1. JournalismThe Guardian2026-10-18

The debate