ChatSOC Invented a Case. The Lawyer Paid for It.

a California workers' comp defense attorney trusted a new chatbot attached to the most trusted source in California workers' compensation. Today's case shows that just because you've always trusted the source, doesn't mean you should trust the source when they add AI. 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.

This is Gilberto Lopez Mendoza vs SWI Finishing, Inc., decided by the California Workers' Compensation Appeals Board on June 26, 2026. Sullivan on Comp isn't some fly-by-night legal app. MS&A published its first edition back in 2011, a multi-volume treatise on California workers' compensation law. By its own account, that treatise became quote, the standard research text on California workers' compensation law, end quote.

Its 16 volumes, updated monthly online, and other releases describe it as the most thorough and up-to-date digital resource in the field, used by more than 8,000 professionals. In April of 2024, Sullivan on Comp launched ChatSOC, like Chat GPT, get it? How clever. They promised it was quote, as easy as chatting with a colleague, end quote. They promised quote, detailed, accurate responses, end quote, while quote, alleviating the burden of time-intensive research, end quote.

So when a defense lawyer needed case law fast, he didn't go around the treatise. He went straight to the tool the treatise built for exactly this. He didn't write the bad law himself. He just signed his name to it. Here's where this one goes.

What the lawyer asked the chat bot and what it handed back. What the judge found. Going looking for those cases directly. What the board said about a lawyer who let software do his legal thinking for him. And what it cost him.

And why it could have cost him a lot more. The underlying fight wasn't even about AI. It was a worker's comp dispute. And a worker's comp judge ruled against the defense. So the defense lawyer filed a petition for reconsideration, asking the board to reverse that call.

Here's the thing about a petition for reconsideration in this system. It has to be verified. Verified means the lawyer is swearing under penalty of perjury that everything in it is true and accurate. To build his legal argument, he researched it using the Sullivan on Comp treatise. And he typed specific questions into ChatSOC.

That SOC answered, and its answers became the case law in his verified petition. Four improper case citations went into that filing. And he hadn't checked them. The judge who'd already ruled against him got the petition and went and checked the cases directly. The first one he cited as Olson vs. Worker's Compensation Appeals Board.

The judge ran it, and that citation doesn't lead to Olson at all. It leads to a completely different case called Brannan, about an unrelated dispute over medical evaluators. The judge searched for an Olson case by name instead, and found one, but it's about permanent disability and apportionment, nothing to do with this fight. In the judge's words, the case, quote, does not exist, end quote. The second one he cited as Reynolds vs. Worker's Compensation Appeals Board.

That case is real, it just doesn't say what he told the board it says. The judge called that citation, quote, improper and a bad faith action, end quote. When the board's notice of intent to sanction came down in December of 2025, it named the employer, the insurance carrier, the claims administrator, the lawyer and his firm, All on the hook for up to $2,500, jointly and severally. Joint and several just means the board could come after any one of them for the whole amount, not just their piece of it. But the employer, the carrier and the administrator hadn't done anything wrong here.

They hadn't touched the research, so the board let them off entirely. The final decision landed on the lawyer and his firm alone. The board wrote that filing a document that misrepresents facts and law, whether or not it was prepared with the assistance of AI, is prima facie evidence of sanctionable conduct. Prima facie is just a fancy way of saying, on its face that's already enough, before anyone even hears an excuse. And then the line that should get framed and hung in every law office in the state, In essence, counsel has delegated legal work in this matter to a non-attorney AI computer program.

Now here's where the sympathy comes in. Because after the fact, this lawyer did almost everything right. He admitted it in writing, in a supplemental petition, laying out exactly what happened. He'd used the treatise, typed questions into ChatSOC, and its answers became the case law he filed. He admitted he failed to exercise an appropriate level of care by assuming the output of chat SOC was accurate.

He apologized, and the board called it genuine. He voluntarily took a continuing education course on the ethical use of AI in legal practice. His firm banned AI and automated tools from supplying legal authority for any filing without independent verification, and rolled out firm-wide training and internal controls to check the existence, accuracy, and applicability of every citation before it goes out the door. Because of all that, the sanction that started as a possible $2,500 came down to $1,000. But the board didn't let it go without a warning.

Quote, in the future, the amount of sanctions imposed may not be so lenient. End quote. The lawyer later inquired with Sullivan on Comp about why ChatSOC had done this. They told him AI systems can generate hallucinations, fabricated information that looks authoritative but is invented. The same word in his own admission to the board.

And here's why that word matters. A hallucination doesn't stammer. It doesn't hedge. You can't hear the difference between the chat bot inventing the law and the chat bot reporting the law. It sounds exactly as confident as real research.

And this one arrived wearing the authority of the treatise California Comp Lawyers Trust Most. He assumed the output was accurate, and signed his name to it. By March of 2026, Sullivan on Comp's own Terms of Service had added that exact hallucinations, plus a duty to check ChatSOC's work. Quote, these inaccuracies are sometimes referred to as hallucinations. Users are responsible for reviewing ChatSOC responses for relevance and accuracy before relying on them.

End quote. Nobody's saying one caused the other. The board never said it, and neither will I. But that's the timeline, in that order. And the board made a point of saying this was never a decision on the merits of who was right.

So here's what actually happened here, stripped down. A tool built to save a lawyer time made him a promise, as easy as chatting with a colleague. But a colleague who cites bad law gets fired. A chat bot that does it gets an updated Terms of Service. The attorney who signs the pleading is still the one who's ultimately responsible for checking the work.

That's not my line, that's the board's. The treatise didn't go to court that day. He did. I'm Lucy Hatchett. Subscribe and I'll see you next time.