Editorial illustration for UC Berkeley Law Bans AI From Graded Work in Skills Debate
AI Ban Fails to Boost Law Students' Reasoning Skills
UC Berkeley Law Bans AI From Graded Work in Skills Debate
Thibault Schrepel expected a ban on AI to sharpen his students' legal reasoning. It didn't. The Vrije Universiteit Amsterdam law professor ran a two-year experiment in his "Law of AI" course, splitting students into three groups to see how access to ChatGPT, or lack of it, shaped their ability to revise provisions of the EU AI Act.
One group worked without any AI tool. A second got AI-generated suggestions with no instruction on how to use them. A third received training in legal prompt engineering, including how to check AI output for accuracy and consistency.
Sixty-six students took part in 2024, and 164 more repeated the setup in 2025. Every student sat the same multiple-choice test and take-home exam, revising a separate provision of the Act under identical grading criteria covering substance, clarity, proportionality, and innovation. Schrepel had assumed the group left to use AI without guidance would perform worse than the group barred from using it at all.
The results, tracked across both years of the course, forced him to rethink that assumption entirely, starting with what happened to the students who had no AI access at all.
One finding held constant across both years: the no-AI group finished last. Banning AI produces worse average outcomes than allowing it, Schrepel concludes.
Why this matters
Berkeley Law and Schrepel are running the same experiment with opposite assumptions, and only one side has two years of data behind it. Schrepel's numbers say the ban itself is the losing strategy: students who got no AI guidance still beat students who got none at all, and structured training beat both. That's a direct challenge to any policy built on the idea that skills have to come before tools.
For founders building AI products aimed at students, this is the argument you want in your pitch deck, an academic willing to say "I was wrong" in print. For researchers, it's a reminder that policy debates in education are running years behind the empirical work. And for anyone at a university weighing a ban, the question isn't whether AI belongs in the classroom, it's whether you're willing to test that belief against a control group before writing it into policy.
Berkeley hasn't published comparable data. Until it does, this is a fight between one professor's study and one law school's assumption.
Related Reading
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Altman, Musk Back Amodei's AI Oversight Call, Says OpenAI Won't IPO
From Archives Anthropic CEO says two factors convinced him to slow AI development
Common Questions Answered
What were the three groups in Professor Schrepel's two-year AI experiment at Vrije Universiteit Amsterdam?
The first group worked without any AI tools, the second group received AI-generated suggestions without instruction on how to use them, and the third group received structured training in legal prompt engineering. All groups were tasked with revising provisions of the EU AI Act to test how AI access affected their legal reasoning abilities.
What did Schrepel's study find about banning AI from law school coursework?
The study found that the no-AI group consistently finished last across both years of the experiment, indicating that banning AI produces worse average outcomes than allowing it. This finding directly challenges policies built on the assumption that students must develop skills before using AI tools.
How did structured legal prompt engineering training compare to other approaches in Schrepel's experiment?
Students who received structured training in legal prompt engineering outperformed both the no-AI group and the group that received AI-generated suggestions without instruction. This suggests that proper guidance on how to use AI tools effectively produces superior results compared to either banning AI or providing it without training.
Why does Schrepel's research matter for AI policy in educational institutions?
Schrepel's two years of empirical data challenge the conventional wisdom that AI should be banned from classrooms to preserve skill development. His findings suggest that the real losing strategy is the ban itself, and that institutions should instead focus on teaching students how to effectively use AI tools rather than restricting access to them.
Further Reading
- UC Berkeley Law School bans most AI use following plagiarism issues - California Courts Newsroom
- UC Berkeley Law Bans AI Use in Coursework, Sparking Debate Over Legal Education - CRBC News
- Berkeley Law's AI Policy: What It Means for AI in Legal Education - Clio
- Law Schools Are Asking the Wrong Question About AI - Inside Higher Ed
- UChicago Law Bans Laptops from 1L Classrooms As Part of Sweeping New AI Strategy for Legal Education - LawNext
