Berkeley tightens – UC Berkeley Law School’s new AI policy, effective this summer, restricts students from using AI for drafting, revising and other exam or writing tasks. But a key policy architect, Chris Hoofnagle, says the school isn’t trying to outlaw the technology—only to p
A UC Berkeley Law student can no longer lean on a chatbot to conceptualize, outline, draft, revise, edit, translate—or use AI for any purpose during an exam once the school’s new policy takes effect this summer.
The change marks a harder line than the law school’s 2023 approach, which allowed students to use AI for brainstorming and conceptualization, such as asking a chatbot to help come up with a paper topic.
Chris Hoofnagle. a UC Berkeley Law School professor who helped develop the new rules. framed the decision as less about banning tools and more about forcing students back toward core lawyering skills. “Our policy is about developing students with the fundamental skills required for AI lawyering,” Hoofnagle told Business Insider.
He said Berkeley recognized the earlier 2023 policy was “too liberal” because generative AI models had advanced since then. With the capability now spreading further, he argued that students could outsource too much work. “It can, in effect, write a research paper soup to nuts,” Hoofnagle said. “So, the increasing capability of LLMs required us to rethink students’ reliance on them.”.
The policy approved a faculty vote, Hoofnagle said, and instructors can deviate from the rules. He also said specific AI-focused courses will follow different standards.
At the center of Hoofnagle’s argument is what he believes first-year law students must learn without AI replacing them: “how to read a case, analyze a case, and write about it cogently.”
“Of course, the question becomes, what is the value add of the lawyer?” Hoofnagle said. “And if that lawyer cannot use their own analytical judgment to assess an AI output, that lawyer has very little value. And so, this is what our policy is about.”
That focus mirrors what Hoofnagle described as growing expectations from law firms—students graduating with proficiency in using AI. “The expectation and demand from law firms is that students graduate with a proficiency in using AI,” he said. “Students are asking for these courses, and they’re learning during their summers that law firms already extensively use AI.”.
Berkeley’s stricter rules also land in a highly competitive market where AI legal tools are chasing growth. Startups like Harvey and Legora are competing for an estimated $1 trillion global legal market. Hoofnagle said Harvey has tried to expand its presence by offering free access to law schools. Stanford Law School—Hoofnagle said had a stricter AI policy when Berkeley rolled out its initial one in 2023—is part of Harvey’s law school alliance program.
Still, Hoofnagle acknowledged that enforcement is inherently messy. “Hoofnagle concedes the policy has loopholes. ” the reporting notes. and he pointed to the reality that policing AI is becoming increasingly difficult as search engines add AI overlays. He said even standard searches on Lexis and Westlaw now include an LLM-generated summary.
“There is no kind of clean answer for it,” Hoofnagle said, adding that Berkeley “obviously can’t ban search.”
The tighter approach fits a broader pattern among schools trying to keep up as AI shifts what cheating looks like and how exams must be protected.
Princeton recently announced what the Daily Princetonian described as the most significant change to its honor code in 133 years. As of July 1, all in-person examinations will be proctored, with the advent of AI cited as one of the reasons for the change.
At Berkeley law. Hoofnagle said the school has seen an “uptick” in misconduct cases and has converted more take-home exams into in-person exams. Those in-person exams are conducted on a computer with specialized software that prevents access to the internet and the use of copy-and-paste functions. Even with precautions, he said it remains possible to cheat.
“You can’t protect against everything in the world, but there’s a really strong signal that if you cut corners, ultimately the student will pay for it when they have to take an in-person exam,” Hoofnagle said.
UC Berkeley Law School AI policy generative AI LLMs academic integrity in-person exams Chris Hoofnagle Harvey Legora Lexis Westlaw Princeton honor code