Fake Citations in Medical Malpractices Cases Are a Supervision Problem

The act of an attorney creating fictitious case law has stopped being a curiosity. Over 1,148 documented instances of hallucinations by attorneys in the U.S. Courts have been identified, and this figure is increasing. Each of these filings had a signature on it. Medical malpractice cases are not exempt. These briefs rest on case law for the standard of care, causation, and expert admissibility, and on medical literature too, so a fabricated source has two places to hide.

Penalties have increased as well. The largest penalty issues to date to attorneys in the United States occurred in Couvrette v. Wisnovsky, an Oregon Federal Court Case. After submitting briefs citing non-existent cases and quoting non-existent language from cases that were never written, the attorneys in this matter received a sanctions award of approximately $110,000. 

This is the highest end of potential sanctions to date, and indicates the extent to which courts may go when a submission is reckless. Malpractice files are already expensive to run because of expert fees and records review, and a sanction on the scale comes on top of the cost.

Signing The Brief Shows Liability

Courts have adopted a straightforward standard, with whoever signs the brief being responsible for the accuracy of all information contained therein. To state that a brief was generated by software is not given any weight by the courts, nor is putting the blame on a colleague or a retained expert. No attorney at the bench cares which tools existed between the attorney and the fictitious citation.

Partners bear the greatest burden. As a general proposition, supervision is a partner’s responsibility. Therefore, when a partner signs off on another attorney’s draft prepared using AI assisted drafting tools, they have committed the same violation as the drafting attorney, with greater culpability due to their higher level of experience. In a medical malpractice case, the same logic reaches the expert’s report. If an affidavit cites a study that doesn’t exist and the attorney files it, the attorney is the one answering the judge.

An attorney who confesses to an error during the initial hearing presents themselves very differently than one who continues to represent that the fictitious citation is valid or attributes the error to the vendor. If an attorney provides a false explanation for the citation it is not merely a mistake on the attorney’s side but rather someone who is trying to mislead the court. This will result in action being taken against the attorney by the state bar disciplinary body.

Paying for Legal AI Does Not Fix the Issue

Many firms believe that by subscribing to a software platform such as Westlaw or Lexis solves the problem. Stanford University conducted a study concerning the use of legal research tools which indicates that hallucination occurs approximately 33% of the time with Westlaw’s AI research and 17% of the time with Lexis + AI. 

It is worth noting that this study was conducted in 2024, and that the tools may have improved since then. Nevertheless, even assuming that rates are reduced, it will take an attorney to ensure that each and every citation is correct.

Fabricated citations appear legitimate because they contain a volume number, reporter designation and page number similar to a legitimate citation. The citation appears within a paragraph that resembles a good associate-authored paragraph. A fabricated citation will only be detected when the author retrieves the case. 

Obtaining the case takes a couple of minutes but reviewers typically do not pull the case because they feel the draft is complete. Medical references fail the same way. An invented article arrives with an author lost, journal name and a DOI-number, and it supports the exact theory the draft needs. Reviewers check a cited study even less often than a cited case, because most lawyers can’t tell by looking whether the study is real.

Use It Where the Source Is On Your Desk

The majority of problems arise from instructing an AI tool to create authority based upon memory. When instructed to create authority from memory, there exists virtually no discernable difference between correct and incorrect output.

Consider a Missouri medical malpractice lawsuit. Early in litigation, the plaintiff will need an affidavit from a qualified expert. The affidavit will require a chronological account constructed from thousands of pages of medical charts, nursing notes, imaging reports, and billing records. Creating a model draft of this chronology makes sense because each entry can be crossed-checked against the original documentation. 

Bad summaries are easily detectable, but fictitious cases referenced in brief are only detectable if someone reviews them. A chronology can also leave out an entry, so the reviewer needs to spot-check for gaps as well as errors, but every one of those checks runs against the chart.

The test is to determine if there is documentation in front of a human that can be compared to the output generated by the tool? If so, then significant hours are saved. If the answer depends on trusting the tools memory from a 1994 case  then it isn’t.

A Rule Worth Writing Down

A one sentence rule regarding drafting requirements is preferable to forty pages of AI policy. A reviewer or associate that is named needs to confirm that the brief is correct, and that every case and medical study cited in or, or in the expert’s report, exists. Exemptions for partners do not apply. Saying you were busy has never been a good answer for a judge and will not be the case here.

The named person must be someone who did not draft the original brief. Authors read their own citation as they meant them while a second reviewer reads what is actually on the page. Include this requirement in your scheduling as well. Verification completed at 11PM just prior to submission deadlines is done badly. Expert disclosure and affidavit deadlines are fixed, so schedule the check for days before them, not the night of.

If anything adverse happens while using an AI tool, documenting precisely what happened will provide value above and beyond the minor administrative costs of having your documents thoroughly reviewed.