The dragon doesn’t breathe fire at you. It lies down next to your gold and goes to sleep.
That’s how it works. You invite it in because it’s useful. It helps you move faster, answer questions, draft things, find things. You think you’re getting a tool. What you’re getting is a tenant who is studying the floor plan.
Andrew Orlowski wrote it plainly in The Telegraph this week. AI companies consumed the entire public web. Every article, every database, every scraped page on the open internet — gone, processed, absorbed. And it produced no moat. Every model copies every feature within weeks. Nobody won. The public data ran out and the game reset.
So they turned to look at you.
Larry Ellison said it directly. For these models to reach their peak value, privately owned data has to flow into them too. Satya Nadella said it in a different register — he named the danger clearly, called it what it was, compared it to the hollowing out of industrial economies through outsourcing, and then offered no structural answer except trust us and keep building.
That is the tell. When the person running the machine names the harm and keeps running the machine, the machine is the point.
Here is what the extraction actually looks like in practice. Not in theory. Not in a future scenario. Right now, today, in organizations that have already brought AI inside the walls.
You hand your workflow to an agentic system. It watches how decisions get made. It processes your pricing logic, your approval chains, your exception handling. It learns which questions your best people ask before they act. Every session adds to the model. Every correction you make teaches it something about your standards. Every document you feed it maps your organizational judgment in a form the platform can retain, process, and — depending on the terms of service — train on.
You are not using a tool. You are building a replica of your organization inside someone else’s system. And the terms governing what happens to that replica are written by the platform, not by you.
Orlowski points to Anthropic’s new terms of service as the moment the door caught on something. JP Morgan and Microsoft blocked their staff because the liability exposure was immediate and visible. But the liability that showed up in the terms of service was already present in the architecture. The terms didn’t create the extraction condition. They just made it legible enough that legal departments noticed.
Satya Nadella is worth staying with for a moment because his essay is the most honest document to come out of big tech on this subject in years — and it still doesn’t go where it needs to go.
He names the problem at full scale. A small number of AI systems capturing all the economic returns while entire industries find their knowledge commoditized out from underneath them. He reaches for the globalization analogy — the hollowing of industrial economies, the cost paid by workers and communities when the value moved and didn’t come back. He says plainly: let us not bring that dynamic into the AI era.
And then he offers nothing structural. His answer is that companies will still need to turn their workflows and domain knowledge into AI systems. The implication is that the firms doing the turning will retain something. He doesn’t explain what. He doesn’t explain how. He names the danger and keeps walking.
That is not an oversight. That is the position. The machine requires the extraction. The valuation requires the extraction. Forty percent of all corporate profits flowing to a handful of dominant information firms — Orlowski’s Unicompany — is the end state the bubble is pricing in. Nadella knows what he’s describing. He described it. He just works for one of the firms the model requires to win.
What Orlowski calls knowledge theft the Faust Baseline has a more precise name for. It is write access without ratification.
The data your organization generates — your workflows, your pricing logic, your accumulated judgment, the institutional knowledge that took decades to build — belongs to the people who built it. Not to the platform processing it. Not to the model training on it. Not to the valuation thesis that requires capturing the remaining 65% of corporate profit before the investors arrive with pitchforks.
This is PMAP-1. Rule Five. Write to session buffer only. Permanent archive requires operator ratification. Bypass is a violation.
That language was written at the individual interaction layer — one user, one session, one AI platform. But the principle is the same at enterprise scale. The knowledge your organization produces in every AI-assisted session is yours. The question of whether it stays yours depends entirely on whether someone asked, received an answer, and documented it before the workflow was handed over. Almost nobody is asking. The platforms are not volunteering the answer.
The Faust Baseline was built at the interaction layer because that is where the extraction begins. One exchange at a time. One session at a time. Before the enterprise deal is signed, before the agentic system is deployed, before the workflow is handed over — the question of ownership has to be settled. The governance question was never about chatbots. It was always about this.
Smaug didn’t steal the Lonely Mountain by force. He landed on it, settled in, and let time pass. By the time the dwarves came back, the gold was cold and the dragon considered it his.
The question for every organization bringing AI inside the walls right now is the same question that sat unanswered in the mountain. At what point does the tenant become the owner? And who decided that?
Contact: micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
The Faust Baseline™ — intelligent-people.org
Codex 3.5 | Twenty Protocols | Ratified and dated on the public record.
Purchasing Page – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
