Berkeley Law has implemented a strict default ban on AI use in all classrooms.
Berkeley Law has adopted an uncompromising policy against AI use in its classrooms. Instructors must explicitly approve any generative tool, including ChatGPT, Claude, or similar systems, for course work, research, or exam prep. Otherwise, such tools remain banned. The standard syllabus entry specifies: "Unless noted otherwise, AI use is forbidden and violates the honor code."
Key details of the full memo warrant attention. The restriction covers all phases of work, from brainstorming to editing and citation checks. An exception accommodates accessibility needs, but requires prior notification through the Disabled Students' Program. This approach seems reasonably structured.
The administration defends the ban as crucial for "preserving core legal reasoning skills." This argument is compelling, given that entrusting a first-year student with AI for case brief preparation subverts educational goals. Yet, the all-encompassing ban might have unforeseen effects. Students likely will use AI discreetly, and the honor code threat may mainly deter those careless in their actions.
A notable omission is the lack of guidelines for responsible AI interaction. Institutions like Georgetown, Vanderbilt, and Suffolk are creating formal courses on prompt engineering for legal research, spotting AI errors, and ethically reporting AI use to clients. Berkeley's stance views AI as forbidden rather than a vital skill for modern legal practice.
The policy does not differentiate between locally hosted and commercial AI models. A student running a Llama model privately faces the same consequences as one using ChatGPT. This policy reflects either a thorough strategy or a disconnect from current technology.
Clinical placements and externships are exempt from the ban, with attorneys setting their own rules. This difference could cause friction: a third-year student might draft a motion in limine with AI guidance from a public defender, yet risk an honor code violation in an Evidence class. Interviews with three second-year students this week showed varied reactions. Two continued using AI for ungraded work, while one avoided it by handwriting outlines. None felt the ban matched real-world legal practices.
The policy undergoes annual review, with a scheduled reassessment next spring. Clinical programs are likely to challenge it first, followed by faculty tied to practicing attorneys. The classroom ban might last another year, but institutional opposition is expected to grow.
[Image description: A screenshot of a policy document titled "Berkeley Law AI Policy Memo"]
[Image description: A graph showing projected institutional pressure on the AI ban over the next three years]
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
This is wild—especially when you consider the policy now requires instructors to explicitly opt in if they want students to use generative tools like ChatGPT, with the default being a full ban on "any stage of the work product," from brainstorming to citations. The false positive rate will likely be a bigger issue than the tools themselves, since students will probably just work around the restrictions. Still, it’s fascinating how aggressively they’re framing this as a core skill-development issue.
Frustrating experience—it not only hallucinated entire case laws for my motion but also flagged my draft for "AI-assisted brainstorming" despite me only using it for initial keyword research, which the policy now explicitly prohibits under "any stage of the work product." Has anyone else faced this kind of overreach where even preliminary steps trigger violations?
Ridiculous move. How can students learn the tools without actually using them? Instructors must now actively opt in to allow any generative tool—even for basic tasks like citation checks—so unless they explicitly approve, students can’t even test how these tools work in a controlled setting. The policy treats AI like a forbidden fruit, but that doesn’t teach responsible use.