Summary
A California Court of Appeal sanctioned an attorney $1,500 and forwarded its opinion to the State Bar after his brief quoted cases that do not exist, quotations generated by AI and never verified. The lawyer had an AI policy requiring human review. The court held the policy itself was not enough, because it never put a competent person on the actual output. If your organization uses generative AI, this is a governance benchmark worth measuring yourself against.
A California Court of Appeal sanctioned an attorney $1,500 and forwarded its opinion to the State Bar after his brief quoted cases that do not exist, quotations generated by AI and never verified. The lawyer had an AI policy requiring human review. The court held the policy itself was not enough, because it never put a competent person on the actual output. If your organization uses generative AI, this is a governance benchmark worth measuring yourself against.
If your business uses generative AI to draft anything that goes out under your name, a California appellate court just showed you how the failure actually happens, and it is not the failure most people brace for. The problem here wasn’t a company that banned AI or ignored the risk. It was a professional who had an AI policy requiring human verification, and who still filed a document full of fabrications. A policy on paper did not save him.
The case is Del Biaggio v. Bansen (Cal. Ct. App., 1st Dist., Div. 4, July 10, 2026), and although it grew out of a dairy contract dispute in Humboldt County, the lesson has nothing to do with cows.
A Dairy Contract That Turned Into an AI Cautionary Tale
Daniel Del Biaggio worked at the Bansens’ dairy under an agreement that promised him salary, livestock transferred over time, and an eventual chance to lease the operation. When the Bansens allegedly failed to hand over the livestock he was owed, he sued and won a $52,850 jury verdict. As the prevailing party, the side that wins, he asked for his attorney fees under the contract, including hours his lawyer’s paralegal had billed. That fee fight is what carried the case up on appeal.
It might have stayed an ordinary fee dispute. Then the court read the opening brief.
What the Court Actually Held
On the money questions, the result was mixed. The court upheld the trial judge’s decision to cut the attorney’s hours, since block billing, lumping many tasks into a single time entry, and several thin causes of action gave the judge room to trim. It reversed the exclusion of paralegal fees, holding that a contract promising the prevailing party its “attorneys’ fees,” a term the contract left undefined, reasonably includes the reasonable cost of paralegal work. It also erased a sanction the trial court had imposed on Del Biaggio’s lawyer, finding that his reconsideration motion, a request that the trial judge revisit its own ruling, was proper.
Then the court turned to something it raised on its own. Del Biaggio’s opening brief attributed quotations to the California Supreme Court’s decision in PLCM Group v. Drexler that appear in no case at all, and it misdescribed other authority. Those invented quotations stated the exact proposition counsel wanted to prove. The lawyer’s own declaration acknowledged the passages came from generative AI and were never properly verified before filing.
Why “We Have an AI Policy” Wasn’t Enough
Here is the part every business should sit with. The lawyer told the court his office uses AI “consistent with State Bar guidance,” with “protocols requiring human verification of all outputs.” The court was unmoved, and its reasoning is the real takeaway.
First, the protocol did not actually deliver the review it promised. The lawyer planned to have a paralegal confirm that the citations existed while he was tied up in another trial. The court said that plan was inappropriate even if it had worked, because he never intended to read the cited law himself. Checking that a source exists is not the same as reading what it says.
Second, the court flagged a detail worth remembering. The fabricated quotations said precisely what the lawyer was hoping to prove. That should have been the warning sign. AI output is most dangerous at the exact moment it tells you what you want to hear.
The sanction was $1,500, paid to the court, with the opinion forwarded to the State Bar. The dollar figure is small. The professional and reputational consequence is not.
What This Means for Businesses Using AI
Read the holding as a governance standard, because that is how it will be used. A written AI policy is not a defense unless the review it promises actually happens, on the real output, performed by someone competent to catch the error. Delegating verification to a proxy step, like confirming a citation exists rather than reading it, is precisely the gap that produced this result. And when AI hands you something that confirms your position perfectly, that is the moment to slow down, not speed up.
For regulated professionals, the exposure is discipline and sanctions. For everyone else, it is contracts built on invented terms, reports with fabricated support, and decisions made on facts that were never real.
Bottom Line
If your organization uses generative AI, the question is not whether you have a policy. It is whether a qualified human reviews the substance of what the tool produces before it leaves the building. This decision turns that from a nice-to-have into a benchmark a court is now willing to enforce. Horst Legal Counsel advises businesses on AI governance, professional risk, and the disputes that follow when controls fail. If you want a candid look at where your current process would leave you, we are glad to talk it through.