Quinteros v. Harbor Distributing is not the flashiest AI citation case on the board. Nobody was fined six figures. Nobody got disbarred. Nobody gave a quote about ChatGPT in open court that will be carved into a law school slide deck. What makes it useful is more mundane and therefore more damning: a law firm filed an opposition brief in a duplicative class action, the brief contained fake legal material, the trial court sanctioned the responsible attorneys, and the appellate court affirmed.

The case began with overlapping wage-and-hour class actions. Lipeles Law Group filed one action in Los Angeles County in June 2024. Later, the same firm filed a substantially similar action in San Francisco County involving Harbor Distributing and related defendants. Harbor moved to stay the San Francisco action under the doctrine of exclusive concurrent jurisdiction, arguing the later-filed case overlapped with the earlier one.

The opposition brief was supposed to explain why the San Francisco case should proceed. Instead, according to the courts, it misstated the law, cited nonexistent authorities, fabricated quotations, and misrepresented cases that did exist. The trial court stayed the action and issued an order to show cause for sanctions. The appellate opinion later described the matter as arising from "evident misuse" of generative artificial intelligence in an otherwise meritless pleading.

The fake law problem

The trial court's concerns were not limited to formatting errors or bad pincites. The opinion says the opposition relied on citations that did not accurately exist, quoted language that did not appear in the cited cases, and misrepresented controlling authority. The trial court counted at least eight fabricated quotations and said that every other purported quotation from a case in the brief was fictitious.

That last detail matters. A hallucinated case is often easy to catch once someone searches for it and finds nothing. A fabricated quote inside a real case is nastier. The case exists, so the citation has a little camouflage. The checker has to open the case and compare the proposition or quotation line by line. It wastes exactly the kind of court and opposing-counsel time that legal citations are supposed to save.

The firm explained that it had used a contract attorney for law-and-motion work. The contract attorney denied using generative AI. He said he used Lexis and Lexis Protege, and that Lexis citation checks had validated the citations. The trial court was not persuaded. It found the explanations lacking, and the appellate court did not find an abuse of discretion in the sanctions.

This is why the story needs careful wording. The court inferred or described misuse of generative AI, but the contract attorney denied using generative AI. That does not make the fake citations harmless. It makes the supervision failure wider. Whether the bad material came from a model, a legal research tool, a bad workflow, or a person wearing a blindfold and typing with hope, the lawyers of record still put it in front of a court.

Delegation is not absolution

The appellate court's most useful point is old law wearing modern shoes: attorneys cannot outsource their duty to verify filings. A lawyer may delegate drafting, research, or cite checking, but the signature on the document still certifies that the legal content has been reasonably investigated.

The court was especially unimpressed with the firm's confusion over its own overlapping cases. The record showed that the firm had filed both actions, yet attorneys offered explanations suggesting they did not fully understand that fact when opposing the stay. One attorney acknowledged not reading the relevant pleadings closely enough. Another conceded that she reviewed the brief for flow but did not fact-check or cite-check it. This is a bad look even before the fake quotations enter the room.

The trial court ordered $6,000 in sanctions: $5,000 payable to Harbor and $1,000 payable to the court. On appeal, the attorneys argued procedural defects and challenged the sanctions. The appellate court held that key procedural objections were forfeited because they were not raised below, and that the remaining challenges failed on the merits.

Damien Charlotin's AI Hallucination Cases Database records Quinteros as a June 11, 2026 California lawyer case involving fabricated case law, false quotations, and misrepresented case law, with monetary sanctions and a bar referral. The database classifies the AI tool as implied rather than specifically identified, which matches the care needed here.

Why this keeps happening

Legal hallucination stories are repetitive because the underlying incentives are repetitive. Lawyers face time pressure. Clients want speed. Courts expect accurate authority. AI tools can produce plausible legal prose much faster than humans can verify it. The mismatch is obvious: the drafting step gets cheaper, but the verification step does not. If anything, verification becomes more important because the output can fail in stranger ways.

The Quinteros brief shows the classic failure pattern. The document apparently looked enough like legal work to be filed. It cited cases. It contained quotes. It advanced a doctrine. But legal work is not typography plus confidence. A citation has to exist, a quote has to appear in the source, and a case has to support the proposition offered. Anything else is theater with a filing fee.

The legal profession does not need another reminder that generative AI can fabricate authorities. It has been receiving those reminders since Mata v. Avianca in 2023, with increasing volume and decreasing judicial patience. Quinteros adds a supervision lesson: if a firm uses contract attorneys, AI tools, citation checkers, or any combination of the above, it still needs a workflow that verifies every authority before filing.

The grimly practical rule is simple. Read the case. Confirm the quote. Confirm the holding. Confirm the jurisdiction. Confirm the procedural posture. Then file. If a tool says the citation is fine but the case text says otherwise, the tool is wrong. Courts do not sanction the software. They sanction the humans who brought it to court.