But let's consider a couple of recent Illinois cases involving AI, Scott v. Illinois Human Rights Commission, 2026 IL App (1st) 251462, and In re Baby Boy, 2025 IL App (4th) 241427. These cases, in very different areas of the law, have only one thing in common: In both cases, the Appellate Court entered sanctions against appellate counsel for submitting briefs that contained entirely hallucinated case or statutory citations, citations to other cases that did exist but did not support the propositions asserted, or quotations that did not actually exist in the cited cases or statutes. Counsel in the Scott case admitted using AI, explaining that he used "a premier corporate subscription of ChatGPT to synthesize complex matters," 2026 IL App (1st) 251462, ¶37 (internal quotation marks omitted). In other words, he paid for the privilege of being sanctioned. Talk about adding insult to (at this point, anyway) unactionable injury....
I don't think it is necessary to get into the particulars of either case; suffice it to say that these two cases involving AI-hallucinated citations are not isolated instances (indeed, in both cases, the reader will find a depressing number of cases, from around the country, in which some of our brother and sister lawyers submitted AI-generated pleadings or briefs without purging the hallucinations therefrom).
Yesterday, I attended a CLE presentation in Judge Alon Stein's Daley Center courtroom on the subject of AI and professionalism (I'd plugged the seminar here). Among the all-star panelists was Appellate Court Justice Rena Marie Van Tine. She offered some seemingly common-sense tips for lawyers seeking to both use AI and stay out of trouble.
For example, she suggested that lawyers using AI must remeber that they -- the lawyers -- are responsible for their AI-assisted submissions, just as they would be responsible for papers prepared with the assistance of a paralegal or associate.
Clearly, this is a correct statement. But an associate, certainly, and a paralegal, one assumes, would know better than to invent a case citation out of whole cloth. A paralegal or associate might misunderstand a holding in a real case, but would not deliberately disregard an unfavorable holding and make up a more favorable one in its place.
So... the attorney who wants to delegate the writing of a brief to AI must be prepared to review every case, every citation, every quote, every asserted fact and record cite, and rewrite and replace every misleading assertion or false statement. Yes, I suppose one must be prepared to do this for every draft produced by an associate as well -- and I have had to do this a few times myself over the years -- rewriting here and revising there and discarding this and adding that, to the point where the final product contains almost nothing of the initial draft. I might as well have written the whole thing myself -- and it might have been done quicker that way, too. But, in these circumstances, no matter how little of the associate's draft survived, I could comfort myself with two things. First, I could use the wholesale changes as a teaching moment, from which the associate would better understand what I wanted to see in future assignments. Second, no matter how much I had to redo, I felt assured that the associate was in no way trying to deceive me or the court. At this time -- with the current state of the art -- there is no basis for such assurance with AI-generated product.
The busy appellate attorney might think -- well, if AI can't be trusted to find and properly evaluate cases or statutes, maybe it could at least summarize a record -- to reduce, for example, a several thousand page trial record, including transcripts and pleadings, into a proper statement of facts for an appellate brief. But even this is apparently beyond the ability of AI at this time: Corporate attorneys have horror stories of AI summarizing contracts or other lengthy agreements and fabricating -- hallucinating -- provisions that simply aren't there (and sometimes ignoring stuff that is in there as well).
The only way AI can be helpful is if you know exactly what it should find, and if you know the subject matter so thoroughly that you can promptly tell that the results produced are indeed accurate. And, if you know all this already, what do you need AI for?
The common thread here is this: AI makes stuff up. When it makes up a picture of the Pope toasting a White Sox victory, that's fine. We know it's made up and we enjoy the joke. When it 'writes' a brief or 'summarizes' a complex document, we may not immediately know what parts of the output are fabricated... but we have more than a fair chance of finding out that some of it has been... because (again) AI makes stuff up.
Maybe -- someday -- maybe someday soon, but maybe not -- AI will advance to the point where it doesn't make stuff up, or only makes stuff up when it's asked to, as in illustration prompts. When that day arrives, if it ever does, humankind will face a watershed moment: All of us, or most of us anyway, will have a skilled lawyer in our pocket, and a genius doctor, gifted engineer, and talented artist besides. We would then have to decide how much control to cede to this new entity -- whether we will struggle to remain in control of it, and build on it and with it, or whether we will let it control us. Maybe we will become a nation of lotus eaters.
Maybe, at that point, humanity will face an existential threat. But not today, folks. Today, AI is merely a threat to our reputations and maybe our licenses. Anyone who uses AI for anything other than silly pictures is jeopardizing their livelihoods. Because AI makes stuff up.
Posts
