UC Berkeley Law bans AI for all graded work starting summer 2026

In early 2025, UC Berkeley Law School announced that starting summer 2026, students will be prohibited from using generative AI for nearly all graded assignments—from brainstorming and drafting to editing and proofreading. Only legal research and professor-approved AI tool training are exempt. Dean Erwin Chemerinsky’s policy is notably strict: the presence of hallucinated citations is treated as prima facie evidence of unauthorized AI use. This is one of the most restrictive and forward-looking bans among top U.S. law schools, set to take effect more than a year after its announcement—a deliberate signal that the institution is drawing a firmer line than its peers.

The landscape of AI policies in legal education reveals a spectrum. Harvard Law School, from fall 2023, adopted a faculty-guided approach that permits AI use with mandatory disclosure and course-specific guidelines. Stanford Law issued interim guidelines in July 2023 that allow AI for brainstorming and editing but prohibit its use for core legal analysis without explicit permission. Yale Law, as of late 2023, left policy to individual professors, creating inconsistency. Berkeley’s uniform ban stands out in its breadth and duration. The business implications are measurable: the legal AI market is already valued at over $3 billion with a 25% CAGR, and RELX (parent of LexisNexis, which owns Casetext) saw a 5% stock dip on news of the ban, as investors weighed potential headwinds from educational adoption. Yet the ban may accelerate demand for auditable AI systems that can transparently trace outputs—a niche that could grow if other institutions follow Berkeley’s lead.

The core tension lies not in whether AI is used, but in how its use is detected and what that detection assumes. Relying on hallucinated citations as a “smoking gun” is unreliable: students can delete generated citations, use models less prone to hallucination, or simply rephrase outputs. More fundamentally, the ban creates a two-tier system: students are still free to use AI for non-graded legal research, potentially giving tech-savvy students an advantage in actual knowledge acquisition while obedient students fall behind. This mirrors a hidden risk that Professor Orin Kerr articulated: the policy tries to preserve foundational skills but may inadvertently penalize those who cannot or do not use AI outside the classroom. The enforcement burden on professors is heavy, and the ban’s prohibition on brainstorming—a human process that AI merely accelerates—feels overbroad and may be challenged as pedagogically unsound. The deeper question is whether legal education’s goal is to produce graduates who can write without AI or graduates who can critically evaluate AI-generated work. Berkeley’s answer is clear, but it may leave students less prepared for a profession where AI-assisted drafting is increasingly standard.

In the end, Berkeley’s move is a corrective to the initial rush of permissive policies, but it risks swinging too far. The most durable path may be one that teaches students to use AI as a junior associate—checking its work, improving its output, and taking responsibility for its errors. Schools that embed AI literacy into core curricula, rather than banning it outright, may produce lawyers better equipped for the practice of the 2030s. Berkeley’s ban will likely push legal tech vendors to build more transparent, auditable tools, but it also buys time for educators to reflect on what truly must remain human in legal reasoning.