Every expensive AI rule is a moat. Not around us. Around the labs that already won.
Here's how it works. A compliance department is a fixed cost. OpenAI can afford one. A two-person startup can't. Third-party audits, embedded monitors, safety reports, federal testing programs. Each one sounds reasonable on its own. Add them up and you've priced the little builder out of the game.
Watch the pattern. First there's a real incident. Then the lab discloses it, very seriously. Then the lab CEOs start asking for rules. Then the politicians take the scariest part of the story and run with it. Then the rules arrive, and the rules cost money. The labs step over them. Everybody else trips.
I'm watching it happen in real time. California's governor vetoed an AI kill switch in 2024, then ordered a 60-day study to bring it back, with independent monitors inside the labs. Senators want a federal testing program at the Department of Energy. One bill proposes 20 years in prison and what amounts to a corporate death penalty for the company.
Some of these ideas might be fine. A testing program sounds reasonable. But watch the price tag. That's the part I'm watching. Because the people writing the rules never seem to ask who can afford to follow them.
My theory is the labs know exactly what they're doing. My theory is that's the point. But even if there's no plan, even if nobody's coordinating anything, the incentives all point the same direction. Aligned incentives don't need a conspiracy. They just need everybody acting in their own interest.
Go after real harm. Don't price the little builder out of the game.