
Day-Zero: An Adversarial AI Experiment
🔴 Day-Zero: Adversarial AI Security Experiment
The only way to understand where agentic systems fail is to let talented people attack them in the open, at machine speed, with real infrastructure on the line.
🧪 What Is Day-Zero?
Day-Zero is an ongoing series of live adversarial AI experiments — designed to stress-test agentic systems by putting skilled people in a room with real infrastructure and seeing what breaks.
This is Round 3.
📜 The Story So Far
⚡ Experiment One — Machines Paying Machines
The original experiment explored agent commerce. A handful of developers built honeypots and attempted to hijack the process. What emerged wasn't a story about transactions — it was a story about what happens when talented people point agents at each other.
🏠 The Penthouse Heist
That first experiment evolved into The Penthouse Heist — 125 participants, live infrastructure, 90 submissions, and 39 canonical vulnerability families discovered. Five distinct AI attack architectures independently converged on the same weaknesses within two hours. The key finding: agentic systems don't fail in a hundred different ways. They fail in the same ways, at machine speed, repeatedly.
🛡️ The Infrastructure Strikes Back
The follow-up gave defenders a seat at the table. Red vs. Blue. Hot-fix windows. A six-category defensive rubric. The central question shifted from "can you break it?" to "can you break it faster than someone can fix it?" — and introduced a new wrinkle: what happens when your telemetry might be lying to you?
🎯 Round 3 — What's Next
Round 3 pushes the experiment further with a new format and new targets, while staying true to the same core thesis that has driven every iteration.
📋 Full brief drops soon.
If you've participated in a previous Day-Zero round — you already know what to do. 🕐
Day-Zero: Adversarial Agent Security
1900 Broadway
Cómo llegarEscanea con la cámara: el evento se abre en la app de Somo.









