Harding spent years as a policy director inside Google DeepMind, so this book has a strange vantage point: someone arguing from inside the institution that most needs to hear what she’s saying. Her claim is simple enough to state in a sentence and still feels almost radical given how AI discourse runs right now. The future of AI is a political choice, not a technical inevitability. The people with the strongest claim to shaping it are democratic publics, not labs, not regulators, not the think tanks the labs fund.

Instead of arguing from AI theory, she builds the case through history: the Space Race, the regulation of IVF, the early internet. IVF is the sharpest of the three. It provoked real public fear and moral panic, and it got governed reasonably well anyway, through actual public deliberation, without a total ban or a free-for-all. Her point isn’t subtle. AI discourse keeps collapsing into “stop everything” versus “ship it and see,” when there’s an entire historical record showing a third option exists. It’s slower and messier. It’s also real.

Where this earns its keep is how much it cuts against the current mainstream. A lot of what’s circulating in AI policy circles right now, RAND’s national security framing, the loss-of-control literature, most of what gets cited on lists like the AI Policy Network’s, treats AI governance as a technical-control problem wrapped in a geopolitical race against China. Harding’s frame is almost the opposite. This is an institutional legitimacy problem: who has standing to decide, and will the process produce something the public will actually live with. Read this against something like the RAND AGI reports and you feel the gap between those two instincts fast.

Here’s where I’d push back. Because she’s writing from inside the industry she’s critiquing, the book stays measured almost to a fault. There’s a lot of space between “publics should have a voice” and any real account of what happens when a lab’s commercial incentives collide head-on with a genuine deliberative process, which is the whole fight. Coming at this from labor organizing rather than policy, the book’s idea of “the public” reads thin. It’s easy to say the public deserves a voice. It’s harder to say how they get one when the entities being governed are also bankrolling a good chunk of the apparatus doing the governing.

Of everything on this list, this is the book that overlaps most directly with the idea that alignment is a governance problem rather than a model problem. Harding is making the institutional-design version of that same argument. It also barely gets cited in labor and worker-power spaces, which is a real gap, because workers are about the most concrete, organized version of “the public” that exists, with real standing and leverage to shape how this technology gets deployed where they work. That connection sits right there in her argument. She never quite makes it herself.

I’d put this on the list for the historical grounding alone. But the real reason to review it publicly is that almost nobody organizing around AI and labor is citing it yet, and they should be.