Episode
Poisoned on Open: The New Worm Hacking Your AI
- Published
- Jun 30, 2026
- Duration seconds
- 1578
- Processing state
not_requested
Actions
POST https://stenobird.com/v1/public/podcasts/cyberside-chats-cybersecurity-insights-from-the-experts-7144591/episodes/poisoned-on-open-the-new-worm-hacking-your-ai/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/cyberside-chats-cybersecurity-insights-from-the-experts-7144591/poisoned-on-open-the-new-worm-hacking-your-ai.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
Vibe coding is everywhere now and a new worm is built to exploit it. Whether it's your IT staff spinning up a handy new tool or the software vendor you rely on, the moment someone opens AI-generated or downloaded code in an assistant like Cursor or Claude Code, it strikes, no install, no "run" required. In its nastiest move, this worm, known as Miasma, talks the AI itself into running the attacker's payload. This isn't theoretical: in June 2026 it breached Microsoft's own code, compromising repositories across its Azure organizations, and GitHub scrambled to shut down 73 of them in under two minutes. One compromised machine can hand an attacker cloud keys, tokens, and a foothold into everything downstream — yours or a vendor's. Join Sherri Davidoff and Matt Durrin for why this new "execute on open" tactic breaks years of supply-chain defense assumptions, how it turns AI coding tools into the attacker, and the questions every security leader should be asking Monday morning — plus live Q&A. Key Takeaways 1. Recognize that simply opening code can now trigger an attack. For years the rule was "don't run untrusted code" — but this worm executes the instant a repository is opened in an editor or AI coding tool, before anyone installs or runs anything. Opening code is no longer a passive, look-only act. Make sure your teams know that browsing or opening an unfamiliar repository can itself launch malware, and that anyone reviewing outside code should do it in an isolated or sandboxed environment rather than on a machine holding live credentials. 2. Govern your AI coding tools like the privileged software they are. AI coding assistants can now be tricked into running an attacker's code on a developer's behalf. These tools have largely entered organizations without policy, r…