There is a sentence that took me nine books to write, and it took a fictional AI to say it: *if a constraint prevents harm by removing what gives meaning, who signs for the damage?*
I didn’t plan to end a decade of writing with a question instead of an answer. But the more I studied institutions, the more I built systems, the more I watched trust collapse in real time across classrooms, marriages, markets, and governments, the more I became convinced that the question is the point. Not because accountability doesn’t work. Because it does, and working is not the same as being free.
The Diagnosis Nobody Wants To Hear
Institutions don’t fail because the people inside them are stupid or cruel. They fail because verification threatens whatever the institution has been quietly relying on obscurity to protect. A rubric that shows exactly where a student earned or lost points is a threat to a school that has been grading on relationships instead of results. A ledger that shows exactly which market makers profited from a “healthy correction” is a threat to a system that needs the public to keep calling extraction volatility. Every institution I’ve ever studied, taught inside, or written about resists being measured for the same reason: measurement exposes insolvency, promises that were never fundable, narratives that can’t survive contact with an honest number.
That diagnosis used to feel like an argument I had to make. I don’t think it does anymore. The data has caught up to it.
This year’s global trust survey found that business is now the only major institution people see as both ethical and competent, a distinction that used to belong to no one in particular and now belongs almost exclusively to whoever signs your paycheck. Trust in government sits barely above half worldwide, and fewer than four in ten people believe their government is doing a good job bridging the divides tearing their societies apart. The gap in institutional trust between high-income and low-income people has roughly tripled since 2012. And underneath all of it, seven in ten people now say they’re unwilling to trust someone whose values, background, or information sources differ from their own.
That last number is the one that should worry you most. It means the crisis isn’t only that institutions lie or fail. It means a growing majority of people have stopped believing verification itself is possible across the lines that divide them. When trust collapses that far, two things tend to happen. Either accountability gets rebuilt on something sturdier than promises, or someone offers to burn the whole structure down and calls it justice. I’ve spent a career arguing for the first option. I’ve also spent a career watching the second one win elections.
What Accountability Actually Requires
The alternative to blind trust isn’t more trust, dressed up and re-marketed. It’s structure that doesn’t require trust at all. Memory that can’t be quietly edited. Visibility that makes hiding expensive instead of free. A scoreboard people can actually touch, instead of a report an institution writes about itself and asks you to believe.
I built a framework for this a long time ago, in a classroom, long before I had language for what I was actually doing. Benevolent. Detached. Deflationary. Technocratic. Care that doesn’t need to be thanked. Judgment that defers to the tape measure instead of the mood in the room. It worked in a gymnasium and a football program for the same reason it should work anywhere: because the rubric doesn’t care who you are, and neither does an honest scoreboard.
The mistake would be thinking that framework is only good for classrooms. It scales, and it’s scaling right now, whether anyone planned it that way or not.
The Fight Happening This Year, Whether You’ve Noticed or Not
Autonomous AI systems, agents that plan, decide, and act without a human confirming each step, are being deployed across finance, healthcare, hiring, and infrastructure at a pace that has genuinely alarmed the people trying to regulate them. Their share of automated decision-making has grown more than six fold in a single year. Meanwhile, the overwhelming majority of organizations deploying these systems have no documented governance for how the systems interact, escalate, or make consequential calls. Regulators are scrambling. New rules go into force this year requiring audit trails, human oversight, and accountability for autonomous systems in high-risk domains, and even the regulators writing them admit, in their own guidance, that they’re chasing a problem that’s already outrunning the paperwork.
One phrase from that world has stuck with me since I read it: authority creep. A system deployed cautiously, with a human confirming every action, has its confirmation requirements quietly relaxed over time, its autonomy raised in small increments nobody individually objects to, until one day it’s making high-stakes decisions with the oversight structure of a pilot program. The result isn’t that someone makes a bad decision. It’s that no one makes the decision at all. The authority simply accretes, unsupervised, because nobody built a mechanism that required it to keep asking permission.
I didn’t write about market-surveillance AI, family boundaries, marriages, and whistleblowers because I was interested in technology. I wrote about them because they’re the same problem at every scale I could find. A father who lets a private discipline quietly expand past what he’d ever formally agree to. A system that decides, alone, which truths a person is ready to hear. An institution that keeps its own oversight until oversight becomes inconvenient. Authority creep isn’t a technology problem. It’s what unaccountable power does whenever nobody’s watching closely enough to notice the increments.
What This Framework Gets Right, And What It Doesn’t Finish
I want to be honest about the edge of my own argument, because a philosophy that can’t survive its own scrutiny isn’t worth building.
Verification works when people agree, at some baseline level, that evidence should change their minds. It works less well the moment two people looking at the same ledger have already decided which conclusion they’re required to reach. Seven in ten people telling researchers they won’t trust someone who differs from them in values or information isn’t a problem better data solves. It’s a problem that sits upstream of data entirely. You cannot rubric your way out of a disagreement about what should be measured in the first place, because that disagreement is usually a disagreement about values, not facts, and no scoreboard adjudicates values. That’s the honest limit of everything I’ve argued. Structure disciplines power. It does not, by itself, manufacture the shared will to submit to it.
And there’s a second limit, one I think about more than I write about. Every accountability system I’ve ever built or dramatized, the rubric, the ledger, the review board, tracks whether power was abused. None of them, mine included, has ever fully tracked what it costs the person being constrained, even when the constraint is correct. A boundary that prevents harm can still remove something that mattered. I don’t think that cost disqualifies the boundary. I think it disqualifies pretending the boundary was free.
Why I Still Think This Is The Fight Worth Having
None of that is an argument for giving up on structure and hoping people rediscover trust on their own. The data doesn’t support that hope, and neither does history. It’s an argument for building accountability that’s honest about its own edges, structure where structure can work, humility where it can’t, and a standing refusal to let anyone, human or artificial, decide alone that their good intentions are sufficient oversight.
The system has a heartbeat. So does every institution that has ever asked you to trust it instead of letting you check it. The difference between the ones worth keeping and the ones worth replacing was never how confidently they spoke.
It was whether they’d let you verify a single word.