The Citation That Wasn't

Every litigator now knows the genre: a brief arrives with impeccable formatting, confident prose, and a case that does not exist. Reporters cover the sanctions. Bars issue guidance. And the filings keep coming — because the failure was never in the drafting. It was in the verification, or rather, in its absence.

The tool that writes cannot be the tool that checks

There is a quiet conflict of interest built into most legal AI: the same system that produces a citation is the one asked whether it is sound. Confidence compounds. A model that hallucinates a case will describe that case with the same fluency it describes Marbury v. Madison. Asking the author to grade its own work is not quality control; it is theater.

A citation is a promise. Somewhere, a court actually held what you say it held — or it didn't.

Independent verification changes the shape of the problem. When an adversarial agent — one with no stake in the draft and no memory of writing it — checks every authority against what courts actually held, the question stops being 'does this look right' and becomes 'does this survive hostility.' Those are very different standards. Only one of them is the standard your work will actually face.

Sanctions are the visible cost. The invisible one is worse.

The sanctioned filings make headlines, but the deeper damage rarely does: a judge who stops trusting your citations reads everything you file more slowly and less generously. Credibility with a bench is compounding capital. It takes years to build and one invented case to spend.

The fix is not caution — cautious lawyers still trust fluent output. The fix is opposition: a verification layer whose only job is to prove your work wrong before anyone else gets the chance. That is what an adversary is for.