Six Lawyers Made the Same Mistake. One Paid Nothing. Two Paid $15,000 Each.

The six

In Gamez v. County of Fresno, in the Eastern District of California, counsel filed nonexistent citations and fabricated quotations. When the court inquired, counsel explained candidly what had happened. No sanction was imposed at all.

In Fletcher v. Experian, the Fifth Circuit found sixteen fabricated quotations and five serious misrepresentations, and imposed $2,500. The court then did something unusual: it said out loud what would have changed the number. Had counsel accepted responsibility and been more forthcoming, it noted, lesser sanctions would likely have followed.

In In re Nwaubani, the Fourth Circuit found three nonexistent cases. Counsel denied using generative AI and offered explanations the court rejected. Public admonishment.

In United States v. Farris, the Sixth Circuit found fabricated quotations attributed to real cases. Counsel had a forty-year clean disciplinary record and answered candidly. The court removed him from the case and denied him all Criminal Justice Act compensation — real consequences — but the candour was recorded as mitigating.

In Fivehouse v. U.S. Department of Defense, in the Eastern District of North Carolina: fabricated quotations from Fourth Circuit opinions. Public reprimand.

And then there is Whiting v. City of Athens, decided by the Sixth Circuit in March 2026 — an appeal that began, improbably, with a fireworks show in Tennessee in 2022.

The court found more than two dozen fake citations, quotations that appeared nowhere in the cases cited, citations that did not support the propositions they were attached to, and a misrepresentation of the district court's own sanctions order.

Then the court issued a show-cause order. And the order is the reason I am writing this essay, because of what it actually asked for. It did not ask whether the lawyers were sorry. It did not ask whether they had meant well. It asked them to explain their cite-checking process, and to disclose whether generative AI or a ghostwriter had produced the brief.

The court asked for the chain of custody.

Counsel declined to answer substantively. They accused the court of harassment instead.

The Sixth Circuit awarded the opposing side its attorneys' fees, doubled the costs, imposed fifteen thousand dollars against each attorney personally, and referred them both to its chief judge. It is, so far, the steepest individual penalty a federal appellate court has imposed in one of these cases.

What was actually punished

Here is the part I want you to sit with.

The court did not punish being wrong. Every one of those six filings was wrong. Being wrong got one set of lawyers nothing and another set fifteen thousand dollars apiece.

The court punished the absence of a file — and then it punished, far more severely, the refusal to admit there had never been one.

This is not a story about artificial intelligence. It is older than that. Substitute a junior associate's memo for the machine, or a number a colleague swore was right, or a market figure someone half-remembered from a conference. The mechanism is unchanged: a source that was persuasive, that could not be argued with, and that no one opened. The machine did not invent this failure. It industrialised it, and it made the failure fluent, which is worse, because fluency is what we have always used as a shortcut for verification.

And notice what the courts are doing about it. Slowly, case by case, across four circuits, they are converging on a rule that has nothing to do with technology: we will grade your decision by the record you can produce for it. Not by the outcome. Not by your sincerity.

A researcher tracking these orders has now documented over a thousand of them. That number is usually reported as a scandal about AI. I read it differently. It is the largest natural experiment I know of in how institutions judge decisions when the decision and the outcome are held apart — and the finding is consistent. What survives scrutiny is not the person who was right. It is the person who can show the work.

Now your turn, and be honest

You have filed something you did not check.

Not a brief, probably. A board paper resting on a number you got from someone who got it from someone. A hiring decision built on a reference call you never actually placed. A budget line carried forward for four years because it was in last year's budget. A vendor renewal approved because it had been approved before.

You did not get caught, and there are only two reasons for that. Either the number happened to be right, or nobody has yet asked you the question the Sixth Circuit asked those lawyers: what, specifically, did you check, and where did it come from?

The lawyer who was sanctioned and the lawyer who was not made the same decision. One of them met a validator. That is the only difference, and it is not a difference in judgement — it is a difference in luck. We grade ourselves and each other by that luck constantly, and it is the most expensive habit I know.

There is a name for the gap between a decision you made and a decision you could defend. I call it decision debt. Like every other kind of debt, it is invisible while nobody calls it in, it compounds quietly, and the call comes at the worst possible moment — a deposition, a board review, a due-diligence request, a Tuesday night when something has gone wrong and someone asks you to explain yourself.

The fix is smaller than you fear, and I want to be precise about what it is not. It is not more caution, and it is not slowing down. It is one habit, and it costs about ninety seconds:

Before you rely on a source, open one of them at random. Not all of them. One. Every time.

Not because the one you open is likely to be wrong. Because the knowledge that you will open one changes what you are willing to pass along — and because a person who opens one at random will, sooner or later, find the thing that was never there. Every single lawyer in those six cases would have been saved by ninety seconds and one click.

And when the decision is consequential, write down the four things you will wish you had written: what you decided, what you relied on, what you rejected and why, and what would have to be true for this to be wrong. One page, before you know how it turns out. That page is not paperwork. It is the only version of you that will be available to testify later, and it will be a calmer and better-informed witness than the one who has to reconstruct it from memory at three in the morning.

The courts have started pricing decision debt. In the first months of 2026 the going rate reached fifteen thousand dollars, plus fees, plus double costs, plus a referral — and the discount for being able to produce the file went all the way down to zero.

Your validator has not arrived yet. That is not the same as not existing.