The Filing That Follows You Forever
2026-09-01
Nearly 2,000 court decisions worldwide now document a lawyer, litigant, or firm relying on AI-generated fiction in a legal filing. The tracker that counts them, maintained by legal researcher Damien Charlotin, added new cases at five to six per day in April. By June, the pace had climbed to nearly eight a day. Two years into the era of generative AI in legal practice, the problem is not slowing down. It is compounding.
The recent entries read like a pattern, not a series of one-offs. On July 28, an Illinois appellate court sanctioned an attorney in Scott v. Illinois Human Rights Commission over a brief containing ten false citations: fabricated statutory quotations, a case that does not exist, and five real cases that did not say what he claimed they said. On August 6, a Connecticut court dealt with the same problem in Elliott v. New York Bariatric Group. A Cook County court fined a firm $59,500 for citing hallucinated cases in a post-trial motion. In Couvrette v. Wisnovsky, a federal court in Oregon imposed roughly $109,700 in combined sanctions and fees — a record. Bar suspensions, once theoretical, have started.
Every story about this problem so far has framed it as a professional-conduct issue: verify your citations, sanction the lawyers who don't, move on. That framing misses the part that should worry corporate affairs teams and firm leadership more than the fines do.
Each of these rulings is now a permanent, timestamped, publicly searchable document. The Illinois opinion, the Connecticut opinion, the Cook County order — they sit in court records and legal databases indefinitely, and legal databases are exactly the kind of source AI systems draw on when a user asks who a lawyer is, what a firm's litigation history looks like, or whether a particular partner is reliable. The sanction was supposed to be the consequence. Increasingly, it is also the input. A single afternoon of unverified drafting becomes a fact that AI systems can retrieve, summarize, and repeat about that person or that firm for years — outliving the case itself, outliving the client relationship, outliving most people's memory that it happened at all.
This is not limited to law firms, and it is not limited to litigation. Deloitte Australia refunded part of a government contract this year after AI-fabricated citations turned up in a report. EY Canada withdrew a study after an outside review found hallucinated references, invented data, and a citation to a McKinsey report that was never written. Any organisation that lets an AI system's unverified output into a document that becomes public — a court filing, a client deliverable, a regulatory submission, a press statement — is doing two things at once. It is creating the immediate professional consequence. And it is creating a durable record that will shape how AI systems describe that organisation to everyone who asks, long after the original mistake is old news internally.
The instinct inside most firms is still to treat this as a drafting-discipline problem: better prompts, a verification step, a policy memo. Those measures matter, but they address the input side. They do nothing about the fact that the sanctions record itself is now part of the permanent, AI-legible account of who you are professionally — and that record accumulates whether or not anyone at the firm is paying attention to it.
The organisations that will manage this well are the ones that start treating their own AI-legible footprint as something to actively monitor, not something to discover after a client, a recruiter, or a journalist asks an AI system why a firm's name turned up next to the word "sanctioned." Knowing what the record already says — and being able to show, precisely and with evidence, when and how it changed — is quickly becoming as important as avoiding the mistake in the first place.