THE TECHNOLOGY BLIND SPOT

Author’s note

I hold a degree from Berkeley, so my read on the law school’s new AI policy may run favorable. Weigh it accordingly. My training is in physics and engineering, and across a long career in enterprise technology the same discipline held. Most problems reduce to a handful of primitives, and there is rarely a need to rebuild the wheel. The problems that defeat every primitive you have are the ones worth real exploration, and even those can usually be bounded by existing principles long enough to design an experiment and test the new idea. The skill underneath all of it is cutting through the chaff, the judgment that separates the common case from the genuinely unique one. If the unique appeared as often as the common, the common would not be common by definition. Knowing which is which is the foundation of any profession. On that ground, I agree with Berkeley Law.

In the summer of 2026, one of the most technology-forward law schools in the country told its students to stop using the most important new tool in the profession. UC Berkeley School of Law adopted a policy, effective that summer, that bars students from using AI to conceptualize, outline, draft, revise, translate, or edit any work submitted for credit, and from touching it at all during an exam. From the outside it looks like a faculty digging in against the future, when it is closer to the opposite. Berkeley is betting that the skill the profession is about to need most is the one AI makes easiest to skip, and that the only place left to build it is a room where the tool is switched off.

The skill is judgment, and judgment has an awkward property. A lawyer builds it by doing the work the machine now does for free.

A first brief is not valuable because it is good, because the first one is usually bad; it is valuable because the lawyer who wrote it learns, in the writing itself, where an argument goes soft, which authority will not bear the weight placed on it, and what a court will not accept. That knowledge does not arrive by reading a finished brief; it arrives by producing a weak one and finding out why. Strip out the drafting and the apprenticeship goes with it. An associate who has only ever edited a model’s output has never made the mistakes that teach a lawyer to spot them. Handed an AI-drafted motion to review, that associate has nothing to review it against. They can check the formatting, but they cannot check the law, because checking the law is the thing they were never made to learn. [See The Primer’s Broken Promise, The Technology Blind Spot (2026).]

This matters because the profession runs on a signature. When a lawyer signs a filing, Rule 11 makes that signature a certification that the lawyer read the document, that the claims are warranted, and that the factual contentions have evidentiary support. The signature is not a claim of authorship. It is an assumption of responsibility for every line, including the lines a model wrote. The certification does not soften when the draft came from a machine. It hardens.

In Johnson v. Dunn, a federal court in the Northern District of Alabama sanctioned attorneys who filed briefs citing cases that did not exist. A model produced the citations. No one verified them. The lawyers signed. Judge Anna Manasco’s order placed the consequence on the names on the filing, not the software. The firm itself avoided sanction because it had an AI policy the court credited; the individuals who signed did not. The model that invented the authority faced nothing, because a model cannot be sanctioned, cannot be disbarred, and cannot be made to answer for anything. Only the human who signed can. That is the whole design. [See When Attorneys Stop Checking AI’s Work, The Technology Blind Spot (2025).]

The bar has already said as much. ABA Formal Opinion 512, issued in 2024, runs the use of generative AI through the duty of competence under Model Rule 1.1, whose eighth comment requires a lawyer to understand the benefits and risks of the technology they use. Competence here is not knowing how to prompt, but knowing whether the output is right, which is a judgment that prompting cannot supply. The same opinion routes AI through Rule 5.3, the duty that governs a lawyer’s responsibility for nonlawyer assistance. A junior lawyer who cannot evaluate the work cannot supervise the tool that produced it, and a senior lawyer who hands the work entirely to the tool has supervised nothing.

Berkeley’s policy is the upstream version of the same duty. The school states its premise without hedging: “thinking remains the sine qua non of good lawyering.” Professor Chris Hoofnagle, who directs the school’s center for law and technology, framed the classroom goal as the best paper the student is capable of producing, not the best paper a model can produce. The policy treats a citation to a source that does not exist as presumptive proof that a student used AI in violation of the rule, which turns the precise failure from the Alabama courtroom into an honor-code tripwire. The bet is that a lawyer who learns to think without the tool can later use the tool well, and that a lawyer who learns with the tool from the first day may never learn to think at all.

The strongest objection comes from inside Berkeley’s own building. Firms now expect graduates to arrive AI-fluent. Students who spend a summer at a firm that runs on these tools come back asking to be taught them, and a school that refuses risks sending its graduates into a market that has already moved. The critics are not wrong about the demand. A graduate who cannot use AI competently in 2027 will be at a disadvantage on day one. Both things hold at once. The demand for fluency is real, and so is the risk that fluency acquired before judgment produces a lawyer who is fast, confident, and unable to tell when the machine is wrong. Berkeley chose the order deliberately, placing judgment first and the tool second. The wager underneath that choice is that judgment transfers to the tool while tool-fluency does not transfer back to judgment.

There is a real limit to the argument, and it is worth stating before a critic does. Excellent lawyers supervise work they no longer perform. An appellate partner who has not drafted a trial brief in a decade can still take one apart, and a general counsel reviews work she would never write herself. They can do it because they built the judgment earlier, by doing the work when it was their turn. The danger is not that experienced lawyers use AI, since they are the people best equipped to catch its errors; the danger is narrower and slower, and it is what happens to the lawyer who never gets the turn, because the turn was handed to a model.

That lawyer inherits a strange position. The Copyright Office will not call the prompter the author of an AI-drafted work, because the expressive choices were the model’s. The court will still hold the signer responsible for every word of it. No authorship credit, full liability. [See The Better Your AI Gets, the Less You Can Own It, The Technology Blind Spot (2026).] A managing partner cannot rewrite the law, but she can decide who drafts. On Thursday, Catherine should name one motion currently slated to start with the model, hand it to a second-year associate to draft cold, and then sit beside that associate and compare the two versions line by line. The point is not the better draft. The point is the associate who, a year from now, can tell which one is wrong.

Berkeley switched the tool off in the one room where switching it off still teaches something. The firms that keep an apprenticeship are making the same wager with their own associates, whether they have named it or not. Nearly everything in a legal document can now be generated. The signature line is the one place a machine cannot stand, and it is worth something only when the name on it belongs to a lawyer who could have written the document and knows when it is wrong.

About the Author

JD Morris is Co-Founder and COO of LexAxiom, an Agentic AI platform for the business of law. Over a 25-year career, he has built and scaled enterprise technology products across Dell, EMC, VMware, and Cisco, including the first exabyte eDiscovery platform. He holds dual MBAs from Columbia Business School (Finance) and UC Berkeley Haas (Marketing), a Master of Legal Studies in Cybersecurity Law from Texas A&M, and a Master of Engineering from George Washington University. He writes The Technology Blind Spot on the intersection of emerging technology and law. Connect with him on LinkedIn at www.linkedin.com/in/jdavidmorris, on X at @JDMorris_LTech, or on Bluesky at @JDMorris-ltech.bsky.social.

References

1. Johnson v. Dunn, 792 F. Supp. 3d 1241 (N.D. Ala. 2025).

2. Fed. R. Civ. P. 11(b).

3. Model Rules of Pro. Conduct r. 1.1 cmt. 8 (Am. Bar Ass’n 2024).

4. Model Rules of Pro. Conduct r. 5.3 (Am. Bar Ass’n 2024).

5. ABA Comm. on Ethics & Pro. Resp., Formal Op. 512 (2024).

6. UC Berkeley Sch. of Law, Artificial Intelligence Policy (eff. Summer 2026), https://www.law.berkeley.edu/academics/registrar/academic-rules/artificial-intelligence-policy/.