His Co-Counsel Sent an AI Brief. He Filed It Without Reading.

The second draft was worse. A lawyer built an outline with AI, then sent it to co-counsel at another firm, never saying where it came from, without checking one line. They filed it: ten pages, twenty-seven citations, two of which didn't exist. A special master flagged two fakes. They refiled the same day. The flagged citations were gone, the rest of the wreckage still in there, and he wouldn't learn the truth for days.

Hey everybody! Thanks for joining us on Sanctioned for AI. My name is Lucy Hatchett. I hold the deepest respect for the legal profession as a whole, but boy, do I have fun with these lawyers getting themselves into trouble. Everyone in this story will tell you the AI is the villain. It isn't. The AI did exactly what AI does: it generated confident, well-formatted fiction.

The humans did the part only humans can do. They filed it. Three lawyers each had a free chance to check the work, and all three spent that chance on trust. Tonight's case is Lacey versus State Farm General Insurance Company, out of federal court in California. A routine discovery fight, a ten-page brief, and a sanctions order that reads like a warning label.

Here are the three things I'm going to show you. First, how an AI-generated outline became a signed federal filing that not one human being read. Second, why the emergency fix both firms rushed out that same day made everything worse. And third, what the court did once it saw the whole picture, including the one call nobody saw coming.

The underlying fight was about as boring as lawsuits get. State Farm was withholding parts of its claims file, saying the material was privileged. The plaintiffs wanted more of it, and wanted the court to review the documents privately to decide who was right. The court had appointed a special master back in January: a retired federal magistrate judge whose whole job was refereeing discovery fights like this one.

After a hearing on April seventh, he ordered both sides to file short supplemental briefs on the privilege question. That's it. That was the assignment. A short brief about a privilege log, the legal equivalent of a permission slip. But the brief the plaintiffs filed didn't come from a lawyer's brain. It came from a chatbot.

Three of them, actually, stacked like a layer cake nobody ordered. Here's how the sausage got made. A lawyer at Ellis George, one of the plaintiffs' firms, fed the research assignment into not one but three generative AI tools: CoCounsel, Westlaw's AI features, and Google Gemini. Out came a preliminary outline. On April ninth, he emailed that outline to his co-counsel over at K&L Gates, the other plaintiffs' firm.

Then he re-forwarded it on April tenth, for good measure. What he didn't do in either email: say the outline came from AI, or check any of it. He just assumed the drafting team would cite-check everything on their end. The K&L Gates lawyers built the brief on top of it, and nobody at either firm cite-checked a single citation.

One of the K&L Gates lawyers later swore, under oath, that he'd relied on the outline believing its sources were, quote, true, accurate, and already cite checked. Another one admitted, also under oath: quote, I did not independently verify those citations. And the best part: the K&L Gates team didn't know the outline came from AI at all.

Nobody told them. And nobody asked. So you've got two law firms, one AI outline, and a grand total of zero humans who verified a single line of it. It's the most expensive game of telephone ever played, except in telephone the message just gets silly. In federal court, the message gets filed. And that is the lesson, sitting right there in the wreckage.

When co-counsel hands you research, the handoff is not the verification. You check it yourself, or you own it. Now let's talk about what was actually in the brief. Because it wasn't just sloppy. It was creative. Ten pages. Twenty-seven legal citations. About nine of them wrong in some way. At least two of the authorities didn't exist at all.

And several of the quotations were, to use the technical term, made up. Take the long block quote from National Steel Products Company versus Superior Court. Looks great on the page. Real case, real citation format. One problem: the quoted text does not exist in that opinion. Then there's Booth versus Allstate Insurance Company. The special master went looking for that decision and could not locate it at all.

And a third citation, Boone versus Vanliner, got flagged too. So this wasn't a typo problem. This was a brief with imaginary case law standing right next to real case law that had been quietly rewritten. On Monday, April fourteenth, around noon, the special master got the brief, went on Westlaw to look up the cases, and couldn't verify them.

So that same day, he emailed both firms and said, in effect, explain yourselves. Later that day, K&L Gates refiled. The two citations he'd flagged? Gone. The remaining AI-generated problems sitting right there in the body of the text? Still there. The cover email said the citations had been, quote, addressed and updated, and described the bad ones as, quote, inadvertently included.

As if the chatbot had tripped and fallen into the brief. And here's the kicker inside the kicker: the special master still didn't know any of this came from AI. He wouldn't find that out until a later order forced everybody to show their work. That's why the second draft was worse. The first draft was an accident.

The second draft was a cover story. But the special master is a writer. And when he finally got the full story, he wrote one of the great judicial takedowns I've ever read. It's linked in the description, and it's worth reading word for word. He started with the AI itself: using generative AI to generate the first draft of the brief was, quote, flat-out wrong.

Sending that draft to co-counsel without disclosing its sketchy AI origins, he said, realistically put those professionals in harm's way. Then he turned to the firms together: they had, quote, collectively acted in a manner that was tantamount to bad faith. And the refiled brief, still carrying AI errors with no disclosure after he'd flagged the problem, was, quote, perhaps most egregious.

In paragraph two, he gave the episode its epitaph: a collective debacle. But the passage that stopped me cold is in the conclusion, where he wrote that the plaintiffs' use of AI had, quote, affirmatively misled him. That he'd read the brief, been persuaded, or at least intrigued, by the authorities cited, and gone to look them up to learn more, only to find they didn't exist.

His word for that was scary: scary because it nearly led to something worse, those bogus authorities making their way into an actual judicial order. The special master almost cited cases that don't exist, in a real order, because two law firms filed a chatbot's homework. So what did he do about it? He struck both supplemental briefs and refused to consider either one.

The plaintiffs got zero discovery relief out of the whole adventure: no bigger privilege log, no forced production, no private review of the documents. Then the money: twenty-six thousand one hundred dollars of his own fees, for untangling the mess. Plus five thousand toward the other side's legal fees, cut down from what they'd actually billed, as an exercise of his discretion.

All in: thirty-one thousand one hundred dollars. Both firms are on the hook jointly and severally, which is the court's way of saying it doesn't care which firm writes the check, as long as the defense gets paid. And then the twist: he didn't sanction a single individual lawyer. Not one. He wrote that their admissions of responsibility had been full, fair, and sincere, and that their apologies were real and profuse.

Piling on, he said, would not serve justice. The client, by the way, had already been told, and isn't paying a cent. And as far as the docket shows, nobody appealed, nobody got it softened, and the order still stands exactly as written. So here's what nobody in this story will say out loud. Nobody got in trouble for using AI.

The AI did its job perfectly. It produced fluent, confident, beautifully formatted fiction, which is what these tools do. The trouble started when three licensed professionals treated a chatbot's first draft like a colleague's finished work. The sanction wasn't for the technology. It was for the signature. And the defense they offered, that the other firm was supposed to check it, isn't a defense.

It's barely a sentence. If this is your kind of cautionary tale, subscribe, because the next one's already in the pipeline. I'm Lucy, this is Sanctioned for AI, and I'll see you next time somebody files something they didn't read. ---