The other day, I gave a presentation to a group of lawyers and legal professionals about AI use and its benefits and risks. As is my practice, I stressed the need for law firms of any and all sizes to have strong AI-use guidelines. I stressed that these guidelines should cover things like which tools are approved for use, the need to protect confidentiality, and the need to verify all citations and other information from any AI output.
And of course, I also talked about the risks of not having an AI use policy. The risk of unauthorized or shadow use of unapproved tools, the risk of confidentiality breaches, and the risk of citations to hallucinated cases and propositions. My standard shtick.
But for some reason, it occurred to me right before this presentation that I should also mention an added reason for robust AI use guidelines. And that is that good guidelines could protect a law firm if someone goes rogue and fails to verify citations before filing them in a court document. I usually don’t talk much about this since it seems rather obvious. But as it turns out, I was prescient to do so.
On November 13, 2026, the Delaware Court of Chancery issued an order sanctioning a lawyer who signed a brief with hallucinated citations. But it exonerated his law firm because it had proper AI use guidelines in place.
The Wrong Way
Here’s what happened. A lawyer submitted fabricated quotations and mischaracterized legal authorities in a brief. After he was outed by the other side, he owned up to it. Sort of. He filed a “corrected” brief. That brief removed quotation marks from the original hallucinated citations. But it retained the inaccurate legal propositions. Good thinking, dude.
But there’s more. The lawyer blamed a paralegal (why do lawyers always blame the para?). Undeterred, he then tried to blame the other side for failing to engage in a meet and confer before it alerted the court. The court wasn’t impressed: “There is nothing to negotiate or compromise when an attorney submits false citations to a tribunal.” Duh.
And all the while he continued to insist (and probably with his outside voice) that the authorities were “real” and “support the arguments advanced.” As if saying it over and over somehow made it so.
Want to incur a judge’s wrath and get sanctioned? Here’s a good roadmap. As might be expected, the court gave short shrift to the arguments. It sanctioned the lawyer under Delaware’s version of Rule 11, but that was not the end of the story. What about the law firm?
The Law Firm’s Role
Delaware’s Rule 11, like its federal counterpart, provides that “absent exceptional circumstances, a law firm must be held jointly responsible for a violation committed by its partner, associate, or employee.”
The court also cited the Delaware Lawyers’ Rules of Professional Conduct, which require law firm leadership to make reasonable efforts to ensure compliance with ethical rules. It also referenced ABA Formal Opinion 512, which calls on firms to establish clear policies on permissible AI use.
That meant the judge needed to look at what the law firm did to prevent these kinds of things from happening. And when she did, guess what? She found the “exceptional circumstances” that relieved it of any joint responsibility for its rogue lawyer. She determined that the firm had undertaken “substantial institutional efforts to prevent the very problem at issue.”
The Right Way
The firm first barred the use of unapproved GenAI tools back in March 2023. In 2024, it formed a working group to evaluate GenAI tools and study the ethical guidance. In July 2025, it adopted a comprehensive GenAI policy. That policy required an experienced lawyer to review GenAI output for accuracy and completeness.
But it wasn’t just the policy. The firm went further. It developed an internal training program addressing hallucinations and the recommendations of Opinion 512. And, said the court, it addressed the professional obligations “concerning competence, confidentiality, meritorious claims, candor, and supervision.” The program was so good that Delaware’s Commission on Continuing Legal Education accredited it for an hour of enhanced ethics credit.
As the court put it:
In short, [the firm] did the right thing. It warned of GenAI risks, established clear verification policies, and trained its personnel through a substantial, multi-layered approach. [The firm] acted reasonably to prevent GenAI misuse and should not be sanctioned for an individual lawyer bypassing those controls.
That’s an important lesson. Having an AI policy isn’t enough. You have to make it real.
Two Legs Of A Three-Legged Stool
The judge’s order and the firm’s actions clearly show that AI use guidelines, to be effective, must be a three-legged stool.
First, the policy itself must be thorough and robust. It must address all the things that the Delaware law firm addressed in its policy. That includes what tools are approved for use, how to protect confidentiality, and, above all, the requirement that lawyers verify the outputs. But that’s only the first leg.
The second leg involves training and education. Without it, the policy’s protections for the firm are moot. Here’s why: lawyers are naturally skeptical and independent. Most of us believe we are special snowflakes as well. So imposing rules on us without explaining the why and without being transparent about them will lead to disaster. Not only will it breed disrespect for the rules and management that impose them, but lawyers will simply look for ways around them.
And there is another reason why education is important. Most lawyers are busy. They are constantly dealing with deadlines and client pressures. So if given the opportunity, they will choose the path of least resistance to get work done. The rules be damned. I recall once upon a time being told that I was to use the firm’s VPN at all times when working remote. The problem was, using that VPN was slow and clunky. It was a pain in the ass. So you can guess what path I would choose, especially when under the gun. It may not have been the best choice, but it’s what happens when policies collide with the realities of getting work done.
The point is if the firm’s approved GenAI tools are slow and clunky, you can rest assured that lawyers will jump to tools that aren’t. That’s why education and training are key. That’s in part why the Delaware judge exonerated the firm. It had provided the training and the why for its rules, even if one rogue lawyer chose not to follow them.
But There’s A Pesky Third Leg
But the AI use guideline stool has a third leg as well. A leg which the Delaware firm also recognized as it evolved its guidelines over the years. AI use guidelines simply aren’t a check-the-box task. To be effective, they have to be evergreen. They have to evolve with the times. If you created guidelines in 2023, for example, it’s unlikely that those guidelines would have addressed the use of AI agents. Yet today, that’s a use that’s getting more and more common.
This means that firms must do something that many don’t like to do. They need to have someone, or perhaps several someones, keep up with AI developments, risks, and the evolving law. That means someone must spend nonbillable time monitoring those developments and ensuring the guidelines are frequently revised.
That brings us to something else a lot of law firms don’t do very well. It’s not enough for someone to recognize the need for guidelines to be revised. The firm must have a mechanism to approve those revisions that doesn’t take six months and involve never-ending partner meetings.
It’s Not A Choice
It’s complicated, no? Drafting guidelines. Training. Keeping up with rapidly occurring developments. All nonbillable. It might be tempting, therefore, for a law firm to throw up in hands and conclude sanctions are going to happen. That they are just a cost of doing business. A cost will be less than the value of the time spent on guidelines.
But in a world where judges are growing increasingly impatient with hallucinations in court filings, that’s dangerous. Not to mention the fact that hallucinations could lead to embarrassment, potential malpractice claims or, at the very least, unhappy clients.
So it’s really not a choice. Law firms absolutely must have AI use guidelines. They absolutely must train partners, associates, paralegals, and legal professionals, and everyone else in the firm, for that matter, about those guidelines and why they’re necessary. They absolutely must keep those guidelines up to date to deal with new developments. They absolutely must have a mechanism for promptly approving needed revisions.
That’s simply a cost of doing business in these days and times, so roll up your sleeves and get to work.
And by the way, the monetary sanctions imposed by the Delaware judge? She made sure the lawyer paid them out of his own pocket.
Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.
The post AI Use Guidelines: It’s Complicated. It’s Also Not A Choice appeared first on Above the Law.
Recent Comments