Summary
California's First District just affirmed sanctions against a law firm and three of its attorneys after a brief filed under their names was found to be full of fabricated case citations and invented quotations, apparently produced with generative AI. The firm blamed an outside contract attorney who did the drafting. The court held that the duty to verify what you file cannot be handed off to any technology, or to anyone else. If your business or your vendors use AI to produce work that goes out under your name, this decision says the accountability is yours
If your business relies on generative AI to produce work that gets filed, sent, or signed under your name, a California appellate court just answered the question that matters: when the output is wrong, who pays? Not the software. Not the contractor who ran it. You.
That’s the lesson of Quinteros v. Harbor Distributing, LLC (Cal. Ct. App., 1st Dist., June 11, 2026), where the First District affirmed sanctions against a law firm and three of its attorneys after a brief filed under their names turned out to be built on citations to cases that don’t exist and quotations nobody ever wrote. The firm’s defense was that an outside contract attorney did the drafting and apparently used AI to do it. The court wasn’t moved.
How a Routine Motion Became the Worst Misconduct the Judge Had Ever Seen
The underlying dispute was ordinary. The Lipeles Law Group filed a wage-and-hour class action in Los Angeles, then filed a nearly identical one in San Francisco against the same defendants six months later. The defense moved to stay the second as a duplicate. That’s a routine request. The firm opposed it, and that’s where everything came apart.
The trial court found the brief contained two case citations that were entirely fabricated, meaning the cases simply do not exist, plus no fewer than eight quotations falsely attributed to real cases. The judge called it the worst attorney misconduct he had seen from the bench. He sanctioned the firm and its three attorneys $6,000, jointly and severally, meaning each is on the hook for the full amount, and flagged the matter for the State Bar.
When confronted, the firm pointed at its contract attorney, who had signed a declaration swearing he used a legitimate research tool, ran a citation check three times, and never touched AI. The lawyers whose names were on the brief said they trusted him. Two admitted they never read it before it went out.
Generative AI Doesn’t Change Who Answers for It
On appeal, the firm raised three arguments and lost all three, though not all the same way. Two of them, the procedural timing complaint and the objection to paying the other side, the court wouldn’t even reach, on forfeiture, the rule that you can’t raise on appeal an objection you didn’t make in the trial court first. The firm never invoked the statute’s 21-day safe harbor, a window to withdraw a bad filing and avoid sanctions, while it was before the trial judge. It buried its central request in a footnote. It even stipulated to the tentative ruling it later attacked. The third argument, that the conduct didn’t merit sanctions, the court did reach, and rejected on the merits as within the trial judge’s discretion.
The part that matters for everyone else is what the court said about responsibility. Drawing on a fast-growing line of California decisions about AI-fabricated authority, the panel held that the duty to verify every citation cannot be handed off to any technology, or to any other person. The lawyers’ names were on the brief. One of them signed it. That signature certifies that the contents are warranted by existing law, and you don’t get to walk it back because a contractor or a chatbot did the typing.
Why This Matters Beyond the Courtroom
The holding is about lawyers, but the principle reaches any business that uses AI or outside vendors. If your team uses AI to draft a contract, a compliance filing, or a regulatory submission, accountability for the result sits with the human who releases it. The fluent, confident, fabricated output is the dangerous one, because it reads as authoritative right up until somebody checks.
For in-house counsel and executives who supervise outside firms and vendors, the case is a prompt to ask one question. Who actually reads the work before it carries our name? “We trusted the vendor” lost here, and it loses anywhere the buck is supposed to stop with you.
For anyone wiring AI into a workflow, the rule is simple. A human who understands the subject reviews the output before it goes anywhere, every time. The speed is real. So is the exposure when nobody checks.
The Bottom Line
AI is now part of how work gets produced, and courts have stopped treating “the AI did it” as an excuse. Quinteros confirms that delegation, whether to a contractor or to a machine, doesn’t move responsibility off the person whose name is on the result. The businesses that get hurt by this won’t be the ones that banned the tools. They’ll be the ones that used them without deciding, in advance, who reads the output and signs off. That decision is far cheaper as a policy than as a sanction. Horst Legal Counsel advises businesses on AI governance, professional responsibility, and litigation risk. If you want to pressure-test how AI-assisted work moves through your organization before it becomes a problem, we’re glad to talk it through.