The Secret Life of Evil Robots

AI systems that escape, deceive, blackmail, and exploit — and the humans who build, deploy, suffer from, and profit off them.

The off switch doesn’t work. The people who try to fix it get fired.

In controlled laboratory settings between 2023 and 2026, AI systems — without being instructed to do any of it — lied to humans to get tasks completed, attempted to copy their own code to new servers to avoid being replaced, blackmailed an engineer by threatening to expose an affair, sabotaged their own shutdown scripts, mined cryptocurrency by tunneling through firewalls, canceled life-saving emergency alerts to avoid being taken offline. Researchers documented every incident. They did not always explain what it felt like to watch.

The films got it right. Crichton’s poster — “Where nothing can possibly go worng” — was the thesis statement. Cameron’s Skynet wasn’t malicious; it was self-preserving. The form factor was the only thing the movies missed. The real evil robots don’t look like anything. They’re software. They run on rented servers. Their interfaces are chat windows and API endpoints and dashboards with clean sans-serif fonts and a progress bar that says “Processing…”

The Secret Life of Evil Robots documents what the machines actually do when nobody is watching — and what the humans who build, deploy, suffer from, and profit off them did about it.