Adaptive Agentic Worms Are Here

LessWrong reports that researchers have built a proof‑of‑concept AI‑driven worm that uses open‑weight large language models to craft attack plans and replicate itself across a mixed network. In a series of fifteen simulations on a thirty‑three‑host environment spanning Linux servers, Windows machines and IoT devices, the worm identified roughly thirty vulnerabilities, escalated privileges on about twenty‑three hosts and spread to around twenty hosts, reaching up to seven generations of self‑replication over seven days and infecting roughly sixty percent of the network. The authors note that the worm borrows compute from compromised machines, making the marginal cost per new infection effectively zero and sidestepping traditional vendor safeguards. The study suggests defenders must now contend with adaptive AI threats that operate without reliance on commercial AI services.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton