On Friday, a Delaware judge fined a senior lawyer $2,500 for filing a brief with fabricated quotations and mischaracterized cases, produced by an AI tool whose output he had not checked. The letter opinion, from Vice Chancellor Lori W. Will, is dated November 13.
The fine is the small part. Under the court's own rule, his firm should have shared the blame. It didn't. The court looked at the law firm AI policy and training in place before the filing, and declined to sanction the firm.
If you run a five-person family law firm, that part is for you. An AI policy used to be something a firm wrote to feel careful. This one was read by a judge.
What happened in the Delaware case?
The case, Leiske v. Kidd, is a dispute over advancement in Delaware's Court of Chancery. It is a long way from family law. On January 22, 2026, the plaintiffs' lawyer filed an answering brief. Five days later, the other side's reply pointed out fabricated quotations, inaccurate citations and legal propositions the cited cases did not support.
What happened next is the part I'd ask every attorney to read. The lawyer told the court that a generative AI tool had been used to revise the brief, and put the miss down to an oversight in a paralegal's cite check. He filed a corrected brief. The correction took the quotation marks off the quotes that did not exist and left the inaccurate descriptions of the cases in place. A day later he wrote again, faulting opposing counsel for not conferring with him before going to the court, and insisting the cases "are real."
The court's answer: "it is no excuse that the cases cited are real decisions."
Later, in an affidavit, he gave a fuller account. He had been leading a pilot of an AI tool the firm had not approved for general use. He used it to help draft and condense the arguments, handed the cite check to a junior attorney and a paralegal, and filed without verifying the authorities himself. He accepted full responsibility and withdrew the blame he had put on others.
The court noted that he had practiced before it "with distinction for years." It also said that using generative AI in legal practice "is neither forbidden nor inherently problematic." The problem was the checking.
Why wasn't the firm sanctioned?
Start with the default. Court of Chancery Rule 11 says that "[a]bsent exceptional circumstances, a law firm must be held jointly responsible" for violations by its partners, associates or employees. The firm's name belongs in the order unless the firm can show why not.
This firm could. The opinion lists what it had done:
In March 2023, it barred the use of unapproved generative AI tools.
In 2024, it formed a working group of directors and professional staff to evaluate tools and study the ethics guidance, including ABA Formal Opinion 512.
In July 2025, its board adopted a written policy. The policy "requires an experienced lawyer to review GenAI output for accuracy and completeness" and bars using AI as a substitute for independent judgment.
It built its own training on hallucinations and on the duties of competence, confidentiality, candor and supervision. The program was accredited for an hour of ethics credit, and more than 200 of the firm's people had completed it by January 2026.
All of that existed before the brief was filed. So the court wrote: "Given the firm's commendable efforts to prevent GenAI misuse, I decline to sanction it." The firm, it said, "should not be sanctioned for an individual lawyer bypassing those controls."
A caution. This is one decision from one trial court, and it binds no family court anywhere. But the test it used is a familiar one. The court looked to the ethics rule on supervision, under which "law firm leadership must make reasonable efforts to ensure that the firm has measures in effect." Your state very likely has its own version of that rule. Opinion 512 puts the same duty in a sentence: "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI."
I think this is where courts are heading. When something goes wrong, the first question will be about the lawyer. The second will be what the firm had in writing, and since when.
What does it mean for the person who checks the cites?
The lawyer's first move was to point at a paralegal. The court did not accept it. Attorneys, it wrote, "routinely and appropriately rely on skilled paralegals and junior attorneys for citechecking." But the duties that come with signing are "personal and nondelegable," and "a senior attorney who signs a brief cannot disclaim responsibility when unverified GenAI output is submitted to the court."
That protects the paralegal, and it leaves a gap a firm should close itself. The opinion doesn't say whether the people doing the cite check knew AI had been used on the draft. Checking a brief a lawyer wrote from the cases is one job. Checking text a machine produced, where a quotation can be invented whole, is a different one. Whoever checks has to know which job they have been given.
What should a small law firm AI policy include?
You don't need a working group or a board. In an ABA Journal column this September, a New Jersey practitioner argued that for solo and small firms, "A one-page written policy is sufficient." I agree. Here is what I'd put on the page, taken from what the Delaware court credited and what it faulted.
The approved tools, by name. Anything not on the list is not used for client work. That includes the tool a partner is trying out. The Delaware brief came out of a pilot.
What stays out of every other tool. Client names, financial records, anything from the file.
Who verifies, and how. Every citation, quotation and factual statement in anything filed or sent is checked against the source. The attorney who signs confirms it was done, or does it.
Tell the checker. Any draft that AI touched is marked that way before it goes to a paralegal or a junior attorney.
Training, with a record. Who was trained, on what, and on which date. The Delaware firm could show a judge all three.
What happens when something slips. Fix every error, tell the court plainly, and don't put it on staff. A correction that only removes the quotation marks makes things worse.
Then date it and have everyone sign it. A policy written after the mistake is an apology. The court gave credit for what existed before the filing.
If you practice in California, some of this stops being optional on January 1. I wrote about that in what SB 574 requires of family law firms.
Where Del fits
Line one of that policy is a list of tools, and every product has to earn its place on it, ours included. Del, our AI teammate for law firms, does the first pass on the matter file and your team reviews it. When Del answers a question, the source document is attached, so the person checking has the page in front of them. Client data is never used to train models, and attorneys stay in charge. You can read more about how Del works as AI for law firms. If you are comparing products, I sorted them by job in AI tools for lawyers.
The lawyer in Delaware had a firm policy, firm training and years of good standing, and the brief still went out unchecked. A policy won't stop every shortcut. It decides who answers for one, and whether your firm's name is in the order.
If you want to see how Del works on a real matter file, request a demo.
This is practical commentary, not legal advice. Court rules and ethics rules on AI differ by state, so check your own state's rules.
Related reading: SB 574: What California's AI Law Requires of Family Law Firms · Hidden Text in a Court Filing Told AI to Take a Side