In an order entered on January 30, 2023 by the U.S. Circuit Court of Appeals, the Court referred attorney Jae Lee, of the New York law firm JSL Law Offices, P.C., to the attorney grievance panel for including a non-existent state court decision in an appeal in the hope of reviving a medical malpractice lawsuit. The Court founds that her conduct “”falls well below the basic obligations of counsel”. A copy of the Order in Park v. Kim, 2nd Circuit, 22-02057, 1/30/24 can be found here.
When ordered to provide a copy of the decision to the Court, Ms. Lee reported she was “unable to furnish a copy of the decision” and admitted that it had bee “suggested” by ChatGBT. In her defense, Ms. Lee asserted that “no bad faith, willfulness, or prejudice towards the opposing party or the judicial system.” The Court was apparently disagreed.
Ms. Lee admitted that she did not read or otherwise confirm the validity of the (non-existent) decision she cited and that she had used ChatGBT when her own research efforts to find a case supporting her position was unsuccessful. The Court concluded that Ms. Lee failed to determine that the argument she made was “legally tenable.”
The referral highlights the risk of using generative AI that may “hallucinate” convincing, but fabricated or inaccurate, responses.
Historically, attorneys have been trained to “Shephardize” the cases they cite. Shephard’s Citation Service was the primary means by which attorneys determined whether cases they intend to rely have been upheld on appeal, cited in subsequent cases, and how subsequent authority has treated the cases. Electronic citation checking has been incorporated into modern legal research such that the task of checking citations is as simple as entering the case name or court citation into the database to confirm the case exists and purportedly stands for the legal position being asserted.
Ethical canons impose, among other things, a duty of competence and prohibitions on conduct involving dishonesty, fraud, deceit or misrepresentation. The regurgitation of generative AI responses in pleadings or motions without verification likely runs afoul of ethical cannons and may result in sanctions or a grievance referral.