At least 12 U.S. law schools adopted new or revised artificial-intelligence policies over the summer, creating a fall semester in which some students must close their laptops while others must complete AI instruction. The changes, documented Monday by Reuters, show legal education moving beyond general warnings about chatbot misuse toward explicit decisions about which intellectual tasks students must perform without machine assistance.
The policies do not converge on a single model. Some schools prohibit AI for brainstorming, editing or classroom use; others permit it for research, criticism and study while barring AI-generated analysis in submitted work. The divergence matters beyond law schools because it exposes a broader higher-education problem: institutions must now define the difference between using AI as a learning aid and allowing it to perform the reasoning an assessment is meant to measure.
Two Problems Produce Different Rules
Law faculties are responding to two risks that point in opposite directions. If students depend on AI before developing independent judgment, they may weaken the reading, recall and reasoning habits that professional education is designed to build. If schools exclude the technology entirely, graduates may enter a profession where AI-assisted research and drafting are increasingly common without understanding the tools’ limitations or ethical constraints.
A public archive assembled by Suffolk University Law School Dean Andrew Perlman illustrates the scale of that tension. Updated Sunday, it catalogs public records from 180 law schools and identifies 38 with some form of mandatory AI curriculum. It separately tracks rules for studying, submitted work, examinations, course disclosure, uploading materials and faculty support, showing that “AI policy” is no longer one decision but a series of decisions about distinct academic activities.
Policies Range From Prohibition to Permission
The University of Georgia’s new strategy combines an analog-classroom default with a requirement that students take at least one AI-related technology course. The design treats unaided classroom reasoning and technological competence as complementary rather than mutually exclusive. Reuters reported that the policy was partly prompted by accounts of students consulting AI when professors used the Socratic method to test real-time analysis.
Berkeley Law takes a more restrictive starting position. Its summer policy bars students by default from using AI to brainstorm paper topics, summarize legal rules for written work, correct grammar or identify repetition, although faculty members may set different rules. Columbia Law School’s August policy permits AI for such tasks as testing arguments, generating hypotheticals, organizing research and correcting language, but requires submitted ideas and analysis to remain the student’s own. Both schools prohibit AI-generated legal reasoning from substituting for the judgment being assessed, even though they draw the boundary at different points.
Professional Duties Raise the Stakes
Law schools face pressures that differ from those confronting many undergraduate programs. Students in clinics may handle confidential client information, while graduates will be responsible for the accuracy of filings and advice. The American Bar Association’s ethics guidance says lawyers using generative AI must consider competence, confidentiality, client communication, supervision, candor to courts and reasonable fees.
That framework helps explain why a school might restrict AI during foundational courses while still requiring instruction before graduation. Professional competence is not demonstrated merely by producing a correct answer; it includes recognizing unreliable output, protecting sensitive information and retaining responsibility for legal judgment. Conversely, students cannot learn those obligations solely through prohibition if the tools will be part of their professional environment.
Assessment Becomes the Central Question
The emerging policies shift attention from detecting AI use to designing assignments that specify what students must learn. An examination may reasonably require unaided recall and analysis, while a separate exercise may evaluate whether a student can interrogate, verify and improve an AI-assisted draft. Those are different competencies, and treating them as interchangeable can make both assessment and enforcement less credible.
Faculty discretion remains central in most of the policies, but it can also produce inconsistency. A student may move among courses where AI is prohibited, conditionally permitted or required, sometimes within the same semester. Clear written rules, disclosure expectations and assignment-specific rationales therefore become part of instructional design rather than an administrative afterthought. Schools will also need evidence that their chosen approach improves learning, because policy counts alone do not show whether students reason more independently or use AI more responsibly.
The Educator's Takeaway
For faculty and academic leaders, the law-school split suggests that a durable AI policy may need at least three separate layers: institution-wide safeguards for integrity and sensitive information, course-level rules tied to learning objectives, and explicit instruction in responsible professional use. A blanket permission or prohibition can obscure the difference between private study, classroom participation, submitted work, examinations and supervised practice.
The next useful evidence will come from outcomes rather than policy language: academic-integrity cases, student performance on unaided assessments, employer evaluations of graduates, and audits of whether AI-assisted work is accurate and confidential. Until those measures are available, the current policies are best understood as competing educational hypotheses. Law schools agree that students need independent judgment and technological competence; they are testing different sequences for teaching both. The broader higher-education lesson is that AI governance is becoming curriculum governance: once institutions decide which tasks a machine may perform, they are also deciding which human capabilities their courses are responsible for developing and demonstrating.
