The AI Worm You Can't Patch (And the One Already on GitHub)

Last week a worm tore through 73 of Microsoft's official GitHub repositories — and it got in by being read by an AI coding assistant. Meanwhile, researchers just built a worm powered by a free, downloadable AI model that took over 75% of a network on its own... and can't be stopped by patching. This is the AI worm. In this video we break down exactly how it works — the live supply-chain attack (Miasma / Hades), the mechanism underneath it all (self-replicating prompt injection / Morris II), the "unpatchable" research worm out of the University of Toronto, and — most importantly — how you actually defend against it. No hype. No doom. Just a clear picture of a genuinely new kind of threat, and the practical habits that beat it. ⏱️ CHAPTERS 0:00 The worm that hit 73 Microsoft repos 0:54 Miasma & Hades: the supply-chain worm 2:24 How a prompt copies itself (Morris II) 3:20 The worm you can't patch (U of T research) 4:07 Reality check: a fire alarm, not a fire 4:34 How to actually defend yourself 6:00 The one firewall that never expires 🛡️ THE DEFENSE, IN ONE BREATH • Treat AI agents like untrusted interns — sandbox them, no live credentials, least privilege • Anything an AI reads is an attack surface — distrust repos, docs, and "paste this into your agent" • The hardest patch is your own assumptions about what counts as "safe" 📌 SOURCES & FURTHER READING • Scientific American — an AI worm that learns as it spreads • University of Toronto — researchers demonstrate an adaptive AI worm • The Hacker News — Miasma supply-chain worm leak (ThreatsDay bulletin) • InfoWorld — "Hades": the malware that lies to AI security agents • Schneier on Security — AI Worm • Cornell Tech — Morris II, the first self-replicating GenAI worm (proof of concept) 🎓 EVOLUGUARD ACADEMY Plain-English security awareness for humans. If this helped, subscribe and share it with the developer in your life who's a little too quick to open a stranger's repo. This video is for education and defense. Nothing here is a how-to — it's a how-to-NOT-get-hit. #AIworm #cybersecurity #promptinjection #AIsecurity #infosec #supplychainattack #Miasma #Morris2