Berkeley's AI Policy Should Worry an AI EdTech Company. Here Is Why It Doesn't
UC Berkeley School of Law has adopted the kind of AI policy that should make an AI education company nervous.
Effective this summer, Berkeley's default rule prohibits students from using AI to conceptualize, outline, draft, revise, translate, or edit any work submitted for credit. It prohibits AI for any purpose in any exam situation. And it bars students from uploading course materials — assignments, readings, slides, class recordings — into generative AI systems at all.
We anticipated policies like this. It is why JurisNote is built the way it is. But before getting to that, the policy deserves a fair reading, including the parts we would have written differently.
What the policy actually says
The rule is a default, not an absolute. Professors can deviate for courses designed to teach AI fluency, or wherever they judge a different rule pedagogically appropriate — but the deviation must be in writing, with notice, and must require students to disclose any authorized AI use. A student unsure whether a use is permitted must ask and receive clarification in writing before proceeding.
The policy is unusually concrete about where the line sits. Its own examples of prohibited conduct include asking AI to brainstorm a paper topic, propose an organizational structure, compose a paragraph summarizing a legal rule, identify repetitive passages that should be cut, or correct grammatical mistakes. Using AI to generate an exam outline whose elements later appear on the exam is prohibited exam use.
There is one carve-out: AI may be used for research on papers, but only for the limited purpose of identifying sources — cases, statutes, secondary materials. Students remain responsible for accuracy, and the policy adds a provision we suspect other schools will copy: citations to sources that do not exist raise a presumption of prohibited AI use. The hallucinated citation is no longer just an embarrassment; it is evidence.
The stated purpose is the same reasoning we examined in the University of Chicago's device policy: thinking remains, in the policy's words, the sine qua non of good lawyering, and courses should develop the cognitive skills that make AI usable — the ability to deploy it strategically, assess its output critically, and uphold ethical obligations while doing so.
Where we disagree
We agree with the purpose. We do not agree with every line.
Consider the example the policy itself gives for prohibited revising: asking an AI tool to identify repetitive passages in a paper that should be cut. The student wrote every word. The tool constructs no argument, supplies no analysis, and adds no content — it points at the student's own prose and says "you said this twice." Treating that the same as asking AI to compose the paragraph collapses a distinction that matters: whether the intellectual work being graded was done by the student.
Or take a use the policy reaches only indirectly: generating practice questions from a student's own outline. Quizzing yourself against material you built is close to the definition of studying. Under the default rule, whether this is permitted depends on whether the student's own notes count as "course materials" that cannot be uploaded — a genuine gray area the written-clarification requirement will have to absorb, one student question at a time.
A policy that prohibits AI from replacing reasoning is right. A policy that also prohibits AI from checking work the student already reasoned through gives up real pedagogical value for the sake of a cleaner line. We understand why a school drafting under time pressure chooses the cleaner line — bright-line rules are easier to administer and harder to lawyer around, and law schools of all places know their students will lawyer around them. But the cost is real.
The problem Berkeley is actually solving
Set the individual line-drawing aside, and the policy is responding to three legitimate needs at once.
Schools need confidence that graded work reflects a student's reasoning — that is the degree's integrity. Professors need control over when AI supports a course's objectives and when it undermines them — a legal writing seminar and an AI-fluency practicum are different rooms. And students need rules clear enough to follow, because "use your judgment" is not a standard anyone can be fairly disciplined under.
Notice what those three needs imply together: the rules will differ. By institution, by professor, by course, by assignment. Berkeley's default differs from Chicago's approach, which differs from schools that have gone the other direction entirely. Within Berkeley, a professor teaching AI fluency will run a different rule than one teaching first-year writing — the policy explicitly provides for it.
Any educational technology that assumes a single answer to "is AI allowed?" is built for a world that no longer exists.
Governed access
That is the design problem JurisNote set out to solve, and it is why Berkeley's policy validates the architecture rather than threatening it.
JurisNote's AI features are optional and granularly controllable. Schools can adopt the core academic and student-support tools — briefing, outlining, spaced repetition, scheduling — without enabling every AI feature, or any AI at all. Where AI is enabled, the boundaries are configurable rather than all-or-nothing.
We call the model governed access: institutions establish the boundaries, professors retain discretion within them, students can see plainly what is permitted, and the technology adapts to the rule rather than daring students to break it.
Candor requires the obvious admission: under Berkeley's default, several of our AI features should be switched off for graded work, and a Berkeley student using JurisNote's AI to polish a seminar paper would be violating the policy. We built the off switch on purpose. A study platform that only works where AI is unrestricted has bet its students' academic standing on a policy environment moving in the opposite direction.
The world this is heading toward
Berkeley drew its line more aggressively than we would have. But it is right about the underlying issue, and its policy is a preview of the equilibrium: AI use in legal education governed by explicit, written, course-level rules, with real consequences for crossing them and real discretion for the professors who own the pedagogy.
In that world, the question for educational technology is not "how much AI can we ship?" It is "can the institution trust what happens when they turn it on — and turn it off?"
That is not a limitation of JurisNote. It is one of the reasons we built it the way we did.
Learn more about the institutional tools we're building on our institutions page.
Sources: UC Berkeley School of Law, Artificial Intelligence Policy, Effective Summer 2026 · Berkeley Law AI Policy: Questions and Answers · JurisNote, A Law School Banned Laptops. Here Is Why We Agree.