American law schools are rewriting classroom and assessment rules around generative AI, with some banning devices from foundational courses while others explicitly teach students how to use AI in legal work.
Reuters reported that at least a dozen U.S. law schools changed AI-related policies during the summer of 2026, as universities try to reconcile two competing demands: preserving students’ independent reasoning while preparing them for legal workplaces where AI is becoming routine.
The debate matters beyond education because law schools are confronting a question facing knowledge-work industries everywhere: which cognitive tasks should software automate, and which should humans continue practising themselves?
Chicago is deliberately putting friction back into learning
The University of Chicago Law School has adopted one of the clearest approaches.
For the 2026–2027 academic year, the school is piloting a policy that prohibits laptops, tablets and phones across required first-year doctrinal courses, subject to limited exceptions. First-year examinations will also be conducted without access to the internet, electronic files or applications.
The school describes its approach as “AI-resilient pedagogy and assessment”, intended to encourage sustained engagement and discourage students from outsourcing intellectual work before they have developed foundational skills.
But Chicago is not rejecting AI entirely.
Its Legal Research and Writing curriculum will treat writing without AI as the foundation while layering AI-assisted research, revision and drafting on top.
That distinction is important: the goal is not to pretend AI does not exist, but to control when students encounter it.
Columbia is taking a more permissive approach
Columbia Law School has adopted a different model.
Its 2026–2027 default policy says AI tools are already embedded in legal research, due diligence, drafting and other professional work, and argues that legal education must prepare students to use them responsibly.
The policy governs examinations, papers, research memoranda, assignments, clinics and journal work, but provides a framework under which instructors can define acceptable uses rather than implementing a blanket classroom ban.
UC Berkeley Law, meanwhile, says future lawyers may need AI fluency but that students must first retain the ability to conceptualise, outline, draft, revise and edit their own work. Its default policy prohibits AI for work submitted for credit and for examinations unless an applicable exception exists.
AI literacy and AI dependence are becoming separate skills
The scale of experimentation is growing.
A public policy archive created by Suffolk Law dean Andrew Perlman now tracks AI policies and teaching strategies across 180 U.S. law schools.
That variation reflects an unresolved tension.
Professional employers increasingly expect graduates to understand AI tools. Yet educational institutions are worried that using those tools too early can prevent students from developing the judgment needed to evaluate their outputs later.
For product teams, the issue resembles automation elsewhere in knowledge work.
The best system is not necessarily the one that removes the most effort.
Sometimes effort is the mechanism through which expertise develops.
Law schools are therefore testing something technology companies may eventually have to confront themselves: software can make a task easier while simultaneously making it harder for a person to learn how that task works.
The emerging challenge is not choosing between humans and AI.
It is deciding when automation improves expertise—and when it quietly prevents expertise from forming.
