ATHENS, Ga. — American law schools are entering the academic year with sharply different answers to the same question — how should students learn to use artificial intelligence without surrendering the legal reasoning they are there to develop?
At least a dozen schools adopted or revised artificial-intelligence policies over the summer. Some have restricted laptops and phones in class; others require students to study AI; several are doing both.
Protecting the thinking process
The University of Georgia now begins with a default ban on classroom devices and requires at least one technology course related to AI. Dean Usha Rodrigues said the policy is meant to preserve the live, uncertain reasoning demanded by the Socratic method.
The University of Chicago barred laptops and phones in required first-year classes, while the University of California, Berkeley adopted a restrictive default policy that limits AI use even for brainstorming and editing. Columbia permits students to use tools to test arguments and solicit criticism, provided the underlying analysis and expression remain their own. Michigan allows AI-assisted brainstorming and research but not drafting or revision.
Nearly all schools prohibit AI in ordinary examinations, according to educators tracking the policies. At least 36 law schools now mandate some AI instruction.
The professional stakes
Lawyers must protect client confidentiality, provide competent advice and verify representations made to courts. Those duties make legal education an unusually consequential laboratory for generative AI.
A blanket ban may preserve foundational skills but leave graduates unprepared for technology used by employers. Unstructured permission carries the opposite risk — students may outsource the very judgment they need to practice law. The emerging model is therefore a staged one: learn to reason independently, then learn to use the tools under professional constraints.
Why law schools face a distinct problem
Generative systems can produce a plausible case summary or draft in seconds, but legal education is designed to expose the reasoning that comes before the answer. Students learn to identify which facts matter, distinguish a holding from dicta, test an analogy and recognize uncertainty. If a tool performs those steps invisibly, a polished submission may conceal a missing skill.
At the same time, refusing to teach the technology would leave graduates unprepared for practice. Law firms already use AI for document review, research, due diligence and first drafts. Competence now includes knowing when a model is useful, how to protect confidential information and how to verify every proposition before it reaches a client or court.
Assessment will change before doctrine does
The clearest consensus concerns examinations. Schools can use secure in-person testing when they need to measure unaided mastery, then design separate assignments in which students must use AI and document the process. Requiring prompts, source checks and a critique of the output makes judgment visible.
Policies also need to distinguish among tasks. Brainstorming an issue list is different from generating a paragraph submitted as the student’s work. Grammar correction is different from rewriting an argument. A single rule covering every use may be simple to announce but difficult to defend educationally.
Equity and privacy questions
Students do not have equal access to paid models, and tools may perform unevenly across languages and disabilities. A course that requires AI should provide a common platform or an equivalent alternative. Schools must also prevent the upload of clinic files, internship material and other confidential information into consumer systems.
Faculty need training of their own. A policy cannot depend on unreliable “AI detectors,” which can misclassify human writing and create due-process problems in misconduct proceedings. Evidence should focus on the student’s work process, drafts and ability to explain the analysis.
The emerging professional curriculum
The strongest approach will combine protected spaces for unaided reasoning with supervised use of modern tools. Students should leave school able to work without AI, work intelligently with it and recognize when professional duties require them to put it aside.
The present patchwork is likely to converge around that model. The institutions that explain the purpose of each restriction — rather than merely policing a forbidden technology — will have the best chance of preserving trust with students and employers.
