A top law school is sharply restricting AI use in graded legal work

A notable policy line in AI-era legal education

One of the clearest institutional pushbacks against generative AI in higher education is coming from legal training. According to a report on policy changes at UC Berkeley Law, the school will prohibit students from using AI in nearly all graded work beginning in summer 2026, while preserving a narrow allowance for research support.

The rule described in the report is sweeping. Students would not be permitted to use AI tools to brainstorm, draft, outline, write, revise, translate, or proofread submitted work. Exams would also be off-limits. The reported exception is research use, such as locating statutes or case law, but students would remain personally responsible for every cited fact.

That last point is central to the school’s rationale. Law is a profession where authority, attribution, and precision are inseparable from competence. If a system fabricates a citation or subtly distorts an argument, the resulting error is not just academic. It cuts directly against the standards lawyers are expected to meet in practice.

Why legal education may be drawing harder boundaries

The policy reflects a broader tension across professional schools. AI tools can accelerate routine tasks and help users produce polished text quickly. But legal education is not designed merely to produce polished text. It is designed to train judgment: how to read closely, reason from precedent, distinguish facts, construct arguments, and detect weaknesses.

That helps explain why the reported Berkeley rule reaches beyond writing assistance to include outlining and brainstorming. Those activities are not peripheral to legal education; they are part of the intellectual formation that law school is supposed to force students through. If outsourced too early, schools risk producing students who can assemble work product without fully understanding how to build it themselves.

The report says the school’s position is that future lawyers need to develop core thinking skills before AI can be used in a truly useful way. That is a significant principle because it treats AI not simply as a classroom integrity issue but as a sequencing issue in professional formation.

What the exception reveals

The limited carveout for research is also revealing. It suggests the school is not rejecting AI categorically. Instead, it is distinguishing between support that helps locate source material and support that starts to perform the student’s analytical labor. In effect, the institution appears to be saying that finding the law is different from thinking through the law.

Even that exception comes with strict accountability. Students remain responsible for every fact they cite, and fabricated citations are reportedly treated as evidence of prohibited AI use. That approach addresses one of the most visible risks associated with current generative systems: their ability to produce authoritative-sounding but false legal references.

In a legal setting, that failure mode is unusually damaging. Courts, clients, and opposing counsel all rely on a chain of verifiable sources. An invented case or misstated authority is not a minor drafting defect. It can undermine credibility at the point where legal work depends most on trust.

A signal for other professional schools

The significance of the policy extends beyond one campus. Law is one of the professions most directly affected by AI because its workflow includes large volumes of text analysis, research, drafting, and revision. That makes it an early test case for how institutions define the boundary between productive assistance and unacceptable substitution.

Other schools are likely to watch closely. Some may adopt similarly restrictive rules for foundational coursework while allowing broader use in advanced classes. Others may move in the opposite direction and integrate AI into assignments under explicit disclosure rules. What makes the Berkeley approach notable is its clarity: it appears to privilege preservation of unaided thinking over near-term efficiency gains.

That stance will almost certainly remain controversial. Supporters of broader AI use will argue that students should learn with the tools they will encounter in practice. But the counterargument is that professional education has always delayed certain shortcuts until mastery exists underneath them. Calculators, templates, and search systems did not eliminate the need to understand the underlying method. The same logic now appears to be shaping AI policy in law.

If the reported rules hold, summer 2026 may mark an important milestone in the next phase of AI governance in education: not a general backlash against the technology, but a more explicit decision about which parts of human thinking institutions still consider nondelegable.

This article is based on reporting by The Decoder. Read the original article.