AI Tops Penn Law Students on Two Open-Scale Final Exams
Penn Law’s AI model Claude Opus 4.7 ranked top 10% on most finals and topped all students on two open-scale exams.
Why it matters: Legal professionals should note how AI’s strong exam performance signals shifts in legal education and reasoning. GCs and legal tech teams must monitor AI’s impact on legal training and operations.
- Claude Opus 4.7 took 11 final law exams at Penn Law in Spring 2026, ranking in the top 10% on 7 of 10 graded exams.
- The AI outperformed every student on two exams graded using uncapped numerical scales, enabling distinct top scores.
- Claude completed each exam twice: once using training data alone, once with course materials like syllabi and outlines.
- Claude finished exams in about 8 minutes, while students invest weeks preparing and hours testing.
In Spring 2026, the University of Pennsylvania Carey Law School tested the AI model Claude Opus 4.7 by having it complete 11 final exams alongside a class of 92 students. The subjects spanned key legal areas including constitutional law and intellectual property, demonstrating the AI’s broad legal knowledge (Axios Philadelphia).
Claude passed all 10 exams that were graded, placing above the median class score in every case. On seven of these, it ranked within the top 10 percent of students. Most notably, it outscored every human student on two exams graded on an uncapped numerical scale, meaning grades were not limited or clustered, allowing clear differentiation of top performance.
To assess robustness, each exam was completed twice: first using only the AI’s existing training data, and then again with supplemental course materials such as syllabi and outlines. Claude consistently finished each exam in roughly 8 minutes, a stark contrast to the weeks of study and multiple hours spent by human students.
Penn Carey Law takes a permissive stance on AI use in exams, allowing faculty discretion rather than imposing outright bans. This approach reflects ongoing debates about AI’s role in legal education and the integrity of assessment practices.
Experts are taking note. Stanford Law Professor Julian Nyarko expressed surprise at the AI’s high performance, suggesting it marks a turning point where AI can match or exceed human legal reasoning on standardized tests.
Similarly, Penn Law Professor R. Polk Wagner described the AI’s top scores as "a frank surprise". He cautioned that while human creativity remains essential, it may play a diminishing role as AI capabilities grow, urging law schools to reconsider curricula and evaluation methods.
For legal practitioners, these findings highlight AI’s rapid advancement in understanding and applying complex legal concepts. GCs and legal operations leaders should prepare for AI’s increasing involvement in legal research, drafting, and possibly aspects of legal reasoning that have traditionally required human expertise.
By the numbers:
- 11 final exams taken — Claude Opus 4.7’s full test set at Penn Law
- 7 of 10 graded exams — AI scored in top 10 percent
- ~8 minutes per exam — AI’s completion time versus weeks for humans
Yes, but: While Claude excelled in exam settings, real-world legal practice demands creativity, client interaction, and ethical judgment beyond standardized testing.
What's next: Law schools are reevaluating policies on AI use for coursework and exams amid evolving capabilities. Expect more studies assessing AI’s role in legal training.