Effective summer 2026, law students at the University of California Berkeley will no longer be allowed to use AI for most class assignments and exams. That’s because their professors kept finding misrepresented or nonexistent cases cited in students’ work. With the new policy comes some significant changes.
It appears to be the strictest AI policy among the T14 law schools.
What Berkeley’s AI Policy Prohibits
UC Berkeley School of Law’s new Artificial Intelligence Policy will begin prohibiting students from using artificial intelligence to prepare class assignments or take exams.
The policy explains that critical thinking skills are, well, critical for law students to develop. And while AI fluency is necessary for future lawyers, the technology isn’t so technically advanced that lawyers can reliably use it without using their critical thinking skills. Berkeley wants students to develop their thinking so that they’re prepared to critically assess what AI tells them and uphold their ethical obligations to their future clients.
To teach students how to think, students are banned from using AI to think for them. They cannot use AI to:
- Brainstorm paper topics or thesis
- Propose the organizational structure of a paper
- Summarize a legal rule
- Revise or edit papers
- Translate papers originally written in another language into English
Since 2024, Berkeley has restricted AI at the law school. Students could use the technology for a few reasons, like correcting grammar. However, it’s clear that students are using generative AI for more than that because professors continue to report citations to cases that just don’t exist. This new policy takes a firmer stance.
There Are Exceptions
There is an exception for law students — when students take courses designed to teach them how to use AI ethically and legally while practicing law. These courses are generally for students who are nearing graduation.
Students can also use AI for identifying sources for their research papers, including cases, statutes and secondary sources. However, students are responsible for the accuracy of their research and citations. So, if AI hallucinates a case, the law student must attempt to read the case, spot the issue and avoid the citation. Berkeley says that professors will assume that AI was used if a law student is caught with a case citation that just doesn’t exist.
Law professors who want to give their students more leeway in the use of artificial intelligence will also be able to permit it. If they plan to let law students use artificial intelligence, they must do so in writing and with appropriate notice. They also must require students to disclose any authorized AI use. If questions arise, students must ask their professors whether the use is okay before taking action.
AI will always be banned for end-of-term exams, no matter what.
AI Use Has Continued to Expand in the Legal Profession
Meanwhile, AI use continues to spread beyond higher education and into the legal profession. A recent worldwide survey of attorneys found that 80% of large law firms were using — or exploring the use of — the technology.
With that expanded use comes expanded oops. Another report found nearly 1400 cases, including 957 in the United States, in which lawyers citied cases that were grossly misrepresented or simply didn’t exist. These cases were presumably suggested by artificial intelligence, which “hallucinated” them. When the judge or opposing counsel looked into the cited case, they discovered the issue.
One example is a patent infringement lawsuit against Overstock.com in which a federal judge issued an order to show cause for using generative AI in the lawyer’s response, which related to a motion to exclude testimony given by an expert witness. Ultimately, the judge ruled that five attorneys were in violation of Fed. R. Civ. P. Rule 11(b) for not making a reasonable inquiry into the results presented by their AI tools before signing and submitting their response to the court. They were not granted leave to amend the reply.
In another case gone wrong, a million-dollar prison defense team in Alabama was also caught using fake citations. So was a New Jersey lawyer who tried to use AI for legal research but ended up citing six fake cases.
The Point of a Legal Education Is to Teach Law Students to Think
So, it’s clear: As much as we hate to admit it, Berkeley is right. Lawyers must think critically when using artificial intelligence. And the point of a legal education is to teach future lawyers how to think. If law students use AI for everything now, they’ll fail to develop the skills that will prevent them from citing fake cases — or making bigger mistakes — in the future.
Einstein famously said, “Education is not the learning of facts, but the training of the mind to think.” (That may be a paraphrasing of some of the larger themes in an address he gave to students at the State University of New York (SUNY) Albany.) For law students, this is especially true. Their future clients deserve lawyers who can think critically rather than parrot AI. And the students themselves deserve the benefit of a legal education that forces them to think critically about the law.
A recent opinion published in The New York Times made an especially great point. Because large language models (LLMs) work by drawing on ideas that have already been shared on the internet, they have the unintended impact of constricting human creativity — especially for students. AI can give the illusion of creativity, but the underlying ideas it presents often converge into a few homogenized categories. One of the things we need most from new lawyers are new ways of looking at case law, and we won’t get that if AI is all they know.
