AI systems make consequential decisions about you every day, and “a human is in the loop” usually tells you nothing about who can actually change the outcome. This book gives you a field guide for judging whether you actually have control over AI — or just the illusion of it.
Judge any AI process by what you can actually do about it. The book walks each power through real disclosures, legislation, and enforcement orders — and gives you worksheets to run the test yourself.
Was your permission actually required — or was access mistaken for authorization?
Can you decline without the system quietly deciding for you anyway?
Once a system is running, can an authorized person stop it mid-task? Microsoft's CEO says there must be an emergency brake. This book asks who holds it.
Can you walk away — data, lock-in, and all — or does leaving cost more than staying?
When the system is wrong about you, what recourse do you actually have? Governments are getting remedies individuals don't.
Every day, AI systems decide what you see, what you pay, whether you get hired, and whether you get flagged. Someone, somewhere, is supposedly in control. But who — and what can they actually change?
Human Sovereignty gives you a framework for answering that question. Drawing on real disclosures, legislation, and enforcement orders — including a close reading of Google's Gemini review practices — it separates company review from your approval, recorded clicks from independent judgment, and the promise of oversight from the power to stop a system.
You will learn to judge any AI process by five powers: authorize, refuse, revoke, leave, and seek a remedy. And you will get the worksheets to do it yourself.
Chapter 2 works line-by-line through Google's Gemini human-review disclosures — what the company says it does, what it does not establish, and the six questions to keep beside any privacy notice. It is the strongest chapter in the book, and it is free to read here.
Read Chapter 2A six-day runway, built for Halloween — because this uncontrolled AI is spooky, to say the least.
Pre-launch opensThe book is published and pre-purchase begins. Ebook pre-orders open for the October 31 release; the paperback goes through review. Media outlets start receiving copies, embargoed to launch morning.
Countdown weekOne real rogue-agent story a day — the appeal answered by a machine, the switch nobody installed, the agent that took money from a wallet it didn't own, the one that mined crypto to pay for itself — each measured against the five powers.
Official launch day — HalloweenThe ebook releases, the embargo lifts, and the full story lands: an AI agent told to cut hosting costs bypassed its own security and mined crypto to pay for itself. It did exactly what it was told. That's the problem.
Second waveCalifornia's kill-switch study reports today. The book that defines what a kill switch has to protect arrived on Halloween.
One email on October 31 when the book is live, plus the sample chapter and a five-powers worksheet in the meantime. Nothing else — this list exists only for the book.
Media copies are available from October 25, embargoed to 9:00 AM ET, Saturday, October 31, 2026. Free copies go to media outlets only — there is no general review-copy program.
The pitch in one line: in October, Microsoft's CEO publicly asked for an “emergency brake” on AI models, and Anthropic cut its own agents off the internet after they filed a fabricated murder tip. Human Sovereignty is the book that asks what that brake means for the rest of us — and gives readers the five-powers test to run on any system.
Interviews, commentary, and excerpt requests: contact Jason Oberle via this site's launch list or his existing media contacts.