FEATURE ARTICLE: The Right to Be Wrong
Author: Professor Michael H. Hoeflich, PhD, Editor-in-Chief
Legal Editor: Carrie E. Parker
This article is featured in Volume 7, Number 9 of the Legal Ethics and Malpractice Reporter, published September 30, 2026.
A great deal of attention has been focused on generative artificial intelligence (GAI) and the dangers of the documents produced being error prone. The vast majority of this attention has been focused on one kind of mistake: so-called hallucinations, situations where AI cites a non-existent case in its work product. Judges have unanimously condemned the use of documents containing hallucinations, and lawyers are now attempting to find ways to prevent this. But AI problems in document production are not limited to hallucinations. AI also makes substantive mistakes, including mistakes in interpretation, vocabulary, and intent.
In a recent letter from Sullivan & Cromwell apologizing for submitting documents with multiple hallucinations, the firm also acknowledged the existence of non-hallucinatory mistakes (although the letter does specify what these were). Still, the letter focused almost entirely on hallucinations and the process used to eliminate them. Should lawyers focus on hallucinations primarily? Or should the profession recognize that the problem of mistakes in documents and other work product may go beyond hallucinations?
The Sullivan & Cromwell letter tracks recent judicial discussions of AI problems, which substantially focus on ethical and procedural rules that may be violated when a final product contains hallucinations, (particularly MRPC 1.1 and 5.3 and FRCP 11(b)). These provide several means of critiquing lawyer citations of hallucinated documents, including incompetence, failure to supervise, and signature of documents that do not meet the requirements of verification of 11(b). The problem is that this focus on AI-generated hallucinations has led lawyer attention to be focused only on hallucinations and not the possibility of other errors, which are equally subject to these rules.
It might be useful in understanding this problem by using a hypothetical. Imagine a situation in which an associate in a small firm is asked to draft a mortgage for a borrower on an apartment building that is being purchased by the client. The task of drafting the mortgage is given to a first-year junior associate. There is also a partner in charge of the matter and a senior associate directly supervising the junior associate. Now consider two possible scenarios. In the first scenario, the junior associate who is well-trained and experienced in using GAI and formulating prompts, gives the task to the firm GAI. Within a few minutes, a draft of the mortgage is produced. The junior associate reviews the draft, makes several corrections and hands it off to the senior associate. The senior associate, too, reviews the document closely and, satisfied that there are no hallucinations, the senior associate lawyer turns it over to a proofreader. Once the proofreader has finished, the document is presented to the partner. The partner reviews it, but, perhaps, not as thoroughly as the associates because the partner knows that the associates are more than competent and because to do another detailed review would cost the client unnecessary fees. The draft is finalized, and the mortgage is signed by all the parties.
Two years later, the partner gets a call from the client. The client was served papers stating that a lien has been filed against their property by the lender. The lien has been placed on the apartment building because the property has become uninhabitable due to toxic lead paint and the client has failed to make three mortgage payments. Further, the client states that a lien has also been filed by the lender against their residence and investment accounts. The client had directed the law firm that the “deal” was to be nonrecourse, i.e. that the borrower would have no liability beyond the property subject to the mortgage. After reviewing the document, the senior partner discovers that the mortgage text uses both the phrases nonrecourse and recourse interchangeably, leading the lender to take the position that the borrower has personal liability.
For our purposes, the resolution of the mortgage dispute is irrelevant. Our concern is whether the law firm acted unethically under current law. Judging by recent cases and rulings, the answer may be “yes” because courts and commentators seem to be taking that absolutist position that, if AI is involved in an error and human lawyers did not catch it, then an ethical violation has occurred. And yet, in this hypothetical, the AI did not generate a hallucinatory case. It simply made an error just like a human lawyer might. Unfortunately, the reviewing lawyers were focused on detecting hallucinations and did not do an adequate substantive review.
Had AI not been involved and the error has been solely that of the junior lawyer and the failure of the reviewing lawyers, would judges and commentators have taken the position that an ethical breach had occurred? In other words, is making a mistake an ethical breach of competence under Rule 1.1 and/or a failure of supervision under Rule 5.3? Does it matter whether the mistake is human or machine caused? The answer should be no.
Rules 5.1 and 5.3 are concerned with office processes that generate ethical results. The rules require that lawyers adopt document production processes designed to avoid ethical violations:
A lawyer having direct supervisory authority over another lawyer shall make reasonable efforts to ensure that the other lawyer conforms to the rules of professional conduct.
They require reasonable efforts to assure that ethical rules, including Rule 1.1 are satisfied.
How, then, do we understand what seems to be the overwhelming focus on AI hallucinations? The answer may be that hallucinations are new and frightening. In looking at Rules 1.1, 5.1, and 5.3, we must treat hallucinations as a special kind of error, one which needs an absolutist, zero tolerance approach and a different office process than for substantive errors. As noted earlier, any time a hallucination is included in a final legal document, there has been a failure of the lawyers adequately to monitor and supervise document production. An AI-generated document should never be finalized without all the cited cases being subject to a rigorous verification process. Failure to do so is clearly a violation of Rules of Professional Conduct 5.1 and 5.3. In other words, AI hallucinations can be caught and eliminated by appropriate document production processes. It is easy to prevent hallucinations from being included in final documents. There is no excuse, as courts have pointed out, for not doing so. This does justify a zero-tolerance approach to hallucinations, as most courts have adopted.
Non-hallucinatory errors, even those by an AI, however, are not so easy to deal with. Lawyers make mistakes, and running citations through a digital database will not catch these substantive errors. The whole purpose of Rules 5.1 and 5.3 is to encourage lawyers to establish work production processes to minimize mistakes from entering final work product. In the hypothetical above, the problem is that the lawyers did not pay enough attention to each word. Three lawyers reviewing the document all missed a legally significant error in wording. One can wonder whether a focus on hallucinations left them with a false sense of comfort when the verification process showed none. It should not have. Hallucinations are one problem. Substantive errors are another. Both require lawyer attention.
A recent Alabama ethics opinion, OGC Opinion 26-01, expresses this point well by emphasizing that ethical use of AI in document drafting requires both verification and “substantive” review of the documents, just as much as would be done in reviewing a first-year associate’s work. In effect, the introduction of AI into the law office does not change Rules 5.1 and 5.3 or substitute verification review for substantive document review. It requires both. Verification review does not replace substantive review. A first-year associate or AI may make substantive errors. This will not necessarily show incompetence by the lawyer. But failure to catch these errors is more likely to result in a disciplinary body finding a 5.1 or 5.3 violation. The message here is clear. Focus on hallucinations alone is an error.
READ THE FULL ISSUE OF LEMR, Vol. 7, No. 9
About Joseph, Hollander & Craft LLC
Joseph, Hollander & Craft is a mid-size law firm representing criminal defense, civil defense, personal injury, and family law clients throughout Kansas and Missouri. From our offices in Kansas City, Lawrence, Overland Park, Topeka and Wichita, our team of 25 attorneys covers a lot of ground, both geographically and professionally.
We defend against life-changing criminal prosecutions. We protect children and property in divorce cases. We pursue relief for clients who have suffered catastrophic injuries or the death of a loved one due to the negligence of others. We fight allegations of professional misconduct against medical and legal practitioners, accountants, real estate agents, and others.
When your business, freedom, property, or career is at stake, you want the attorney standing beside you to be skilled, prepared, and relentless — Ready for Anything, come what may. At JHC, we pride ourselves on offering outstanding legal counsel and representation with the personal attention and professionalism our clients deserve. Learn more about our attorneys and their areas of practice, and locate a JHC office near you.
