Judicature International (2026) | An online-only publication
Pictured Above: U.S. District Judge Michael M. Baylson (center) and U.S. Magistrate Judge Caroline Goldner Cinquanto (second from right) with University of Botswana law students during a trial advocacy program in April 2026.
Two federal judges used a complex record of the Grenfell Tower fire litigation to create a hands-on trial-advocacy exercise for University of Botswana law students, with generative AI helping create the hypothetical case and deposition materials.
In April 2026, Eastern District of Pennsylvania Senior Judge Michael M. Baylson and U.S. Magistrate Judge Caroline Goldner Cinquanto traveled to the University of Botswana at the invitation of the International Law Institute (ILI) to teach trial advocacy. Working with law clerks Caroline Kam, Madelaine Pisani, and Sarah Goodman, the judges designed a hands-on program in which students practiced opening statements, closing arguments, direct examination, and cross-examination. To create a realistic case for those exercises, the group used generative artificial intelligence (AI) to adapt the complex record of London’s tragic Grenfell Tower fire into a hypothetical civil lawsuit that students could investigate, argue, and try in a mock setting. This article describes the experience of teaching courtroom skills across cultures, the challenges of engaging a large student audience and the potential — and limitations — of generative AI as a tool for legal education.1
Grenfell Tower Fire Litigation in the United States
On June 14, 2017, the Grenfell Tower, a high-rise apartment building in West London, became engulfed in a devastating fire that resulted in 72 deaths and injuries to hundreds of tenants. Shortly thereafter, plaintiffs, represented by a major personal injury firm in Philadelphia, initiated a lawsuit in the Philadelphia Court of Common Pleas. The defendants were Arconic, Whirlpool, and Celotex. The defendants removed the case to the Eastern District of Pennsylvania, and it was assigned to Judge Baylson.
The Grenfell Tower fire was a major calamity, and all plaintiffs and defendants needed to be well represented in every respect. I undertook an active case management role and promptly issued a scheduling order, which gave the plaintiffs the opportunity to request and receive documents and take depositions from officers and employees of the defendants as well as third parties.
In doing so, I drew on my significant experience in cross-border discovery after learning that some components of the Grenfell Tower apartment building and its construction had been designed and/or manufactured abroad. I knew from speaking at many European conferences organized by the Sedona Conference that the extensive discovery process we have in the United States is either unknown or followed in a very limited way in foreign countries — particularly in France, which has a blocking statute to prevent facts that may embarrass French companies from getting into the public record. My initial order on this challenge was to appoint Noëlle Lenoir, a retired judge of the French Constitutional Court, to act as a special master in supervising discovery in France.
I entered orders requiring the foreign entities to produce documents to Lenoir, as they would in the United States to a court-appointed master. Although I wanted to respect French laws about limited discovery, I also wanted to give the plaintiffs an opportunity to secure evidence from these defendants’ records that might shed light on the cause of the fire.
Numerous motions to compel and for protective orders were filed during the yearlong pretrial proceedings. At some point, the defendants moved to dismiss the case on forum non conveniens grounds, arguing that the litigation should proceed in London. After full briefing, I granted the motion.2 The plaintiffs appealed, but the Third Circuit affirmed my decision in a nonprecedential opinion.3
The British government established a public inquiry to review the tragedy, which produced two multi-volume reports of evidence, findings, and recommendations. The litigation ultimately resulted in a monetary settlement involving approximately 900 victims of the fire, including injured tenants and representatives of the deceased.
Using Generative AI to Develop the Hypothetical Case
Judge Cinquanto and I were invited to teach at the University of Botswana Department of Law in April 2026. We agreed that the Grenfell Tower fire litigation would be an engaging and factually rich legal case for the students. We also thought generative AI could help us develop teaching materials for our curriculum.
To create hypothetical deposition transcripts for the trial advocacy simulation, we first identifiedhough ChatGPT has access to a vast dataset of publicly available information and content provided by third-party partners, we chose to upload a curated set ofating the fictional deposition transcripts
We first uploaded excerpts from the Grenfell Tower Inquiry reports. The inquiry was a major UK public investigation that examined the failures contributing to the tragedy. It heard extensive testimony and evidence over more than 400 days concerning building design, fire safety practices, emergency response procedures, governmental oversight, construction materials, and communications among various entities involved in the building’s renovation and management. The resulting reports span more than 2,500 pages. Because of the sheer volume of material, we selected approximately 30 pages from the executive summaries that identified the principal actors and major conclusions reached by the inquiry. In addition to inquiry excerpts, we uploaded a short documentary-style video and several news articles about the disaster. Importantly, all source materials used in the project were publicly available and nonconfidential.
When uploading the materials to ChatGPT, we provided the model with an overview of the educational purpose of the exercise. We explained that we were designing a law school simulation based on the Grenfell Tower fire to help students practice complex civil litigation strategy and trial advocacy. We instructed the model that students would be asked to analyze potential plaintiffs, defendants, deponents, and witnesses in a lawsuit about the fire. Students would then develop a litigation strategy for both sides of the dispute and ultimately present mock opening statements, direct and cross-examinations, and closing statements in a simulated courtroom exercise. We explained that the purpose of creating the hypothetical depositions was not to recreate the exact historical record but instead to generate plausible factual scenarios that students could use to craft litigation strategy.
We decided to develop the deposition transcripts individually, with each focused on a different perspective related to the tragedy. We asked ChatGPT to generate hypothetical depositions for several categories of witnesses, including:
- Representatives of deceased tenants
- Injured tenants
- Firefighters who responded to the scene
- Corporate representatives of the various entities allegedly involved in the building’s renovation or maintenance
- Government employees responsible for fire safety in high-rise residential buildings
The prompts instructed ChatGPT to create a realistic deposition exchange between examining counsel and the witness, while incorporating issues commonly explored in complex civil litigation. The initial outputs generally produced fairly superficial examinations. However, the model was quite adept at identifying the major factual topics and even theories of liability relevant to each witness. We refined these drafts through iterative prompting. Follow-up prompts asked the model to add sections on a topic initially unaddressed or to go into further detail with certain lines of questioning.
Because we had uploaded excerpts of the inquiry report for context, the transcripts would sometimes read as if a given defendant’s liability was a foregone conclusion. To fix this, we prompted ChatGPT to rewrite the transcript as if liability had not yet been determined. The model responded effectively when we explained that the hypothetical depositions would be taking place years before the investigation was completed and the inquiry reports were published. This process ultimately produced more than 60 pages of hypothetical deposition questions and answers.
Using the Hypothetical Deposition Transcripts With Students in Botswana
We appreciated the invitation to teach in Botswana through the ILI, as we both have done previously, but we did not know what to expect in terms of class structure or faculty support. Upon arrival, we were advised that the school had decided to suspend regular classes for the entire week we were there and that we should expect 250 students as our audience.
Law students gather at the University of Botswana for a trial advocacy program led by two U.S. federal judges in April 2026.
First, let us share a few words about the University of Botswana. The school appears to be a very well-run university with many departments. The law school, however, is somewhat separate. For law students there, the first three years consist of general liberal arts education, while the last two focus on the law, with professors mostly lecturing; student participation is relatively limited.
We decided that since teaching trial advocacy was our goal, student participation was essential. At our first meeting, we gave them a very general background about us and the Grenfell Tower fire litigation.
Judges Michael M. Baylson and Caroline Goldner Cinquanto speak with a participant during their trial advocacy program at the University of Botswana.
We thought the best course of action was to divide this large group of students into several subgroups. It turned out we did not have the full group of 250 at any single instruction session. Instead, we had a designated subgroup of about 50–60 students at a time, which we then subdivided into groups of 15–20 students, with one of us taking responsibility for instructing each smaller group.
The first exercise required the students to compose an opening statement for a lawyer representing the plaintiffs. We discovered that some students were very shy and unwilling to raise their hands or participate in any way. Fortunately, we also noted that each group included somewhat older students willing to speak to the group and give examples of an opening statement for a hypothetical trial concerning the defendants’ liability for the injuries and deaths suffered in the fire. This topic took the entire first day as different panel members made their opening statements.
The second day focused on closing arguments, and each of us gave an example of how one might be presented for a product liability case in a U.S. courtroom. When we asked students to take up the topic and present their own version of a closing argument, however, we ran into the same issues as on the first day, as some declined to participate. Others, though, did take part and presented closing arguments to the entire group.
The third day was devoted to having direct and cross-examinations of various witnesses who were hypothetically called to testify. We used the following assumed identities:
- James Carter: Celotex corporate representative
- Robert Black: representative of the Kensington and Chelsea Tenant Management Organization, basically the “landlord” of the apartment building
- Mark Harris: representative of Rydon, principal contractor of the Grenfell Tower Refurbishment Project
On the final day, we summarized the proceedings and requested feedback from the students, which was generally positive. We also met with university officials and the dean of the law school, who said they were impressed with the novelty of this type of instruction in Botswana.
Conclusion
While this “experiment” took only four and a half days, it provided law students abroad a large amount of information and hands-on experience in trial advocacy — at much lower cost than requiring them to purchase textbooks on the same topic.
The project also demonstrates several capabilities of large language models in legal education. The exercise highlighted how instructors can use generative AI to create realistic simulations for students to hone their analytical skills. ChatGPT proved very useful in synthesizing a lengthy and complex factual record and substantially accelerated the drafting process by producing preliminary deposition transcripts that could then be improved through further prompting.
Michael M. Baylson is a senior judge for the U.S. District Court for the Eastern District of Pennsylvania. He is a former U.S. Attorney and has taught litigation at the University of Pennsylvania Carey Law School and internationally.
Caroline Goldner Cinquanto, U.S. magistrate judge for the Eastern District of Pennsylvania, teaches trial advocacy at Villanova Law and held faculty and leadership roles at Temple Law. An international educator and lecturer, she is a fellow of the American College of Trial Lawyers.
Sarah Goodman is a 2023 graduate of the University of Pennsylvania Carey Law School.
Caroline Kam is a 2025 graduate of the University of Pennsylvania Carey Law School.
Madelaine Pisani is a 2024 graduate of the University of Virginia School of Law.
- This article is written primarily in Judge Baylson’s voice, using the first-person singular to recount his experience presiding over the Grenfell Tower fire litigation. The judge-authors use the first person plural when describing their work together in Botswana.
- Behrens v. Arconic Inc., 487 F. Supp. 3d 283 (E.D. Pa. 2020).
- Behrens v. Arconic Inc., 2022 WL 2593520 (3d Cir. July 8, 2022), cert.denied, Behrens v. Arconic Inc., 143 S.Ct. 787 (2023).
