The cases were not weak. They were not there. Lawyers defending State Farm in a Los Angeles fire-damage lawsuit filed pretrial motions containing citations to decisions that do not exist, quotes nobody ever wrote, and holdings no court ever issued. The plaintiff’s team found them while preparing for trial, and the defence firm has now admitted the errors came from artificial intelligence software marketed to lawyers.
The underlying dispute is ordinary. Fa’alagilagi Meni-Siliga sued in July 2024 over fire damage to her home, naming a contractor, the contractor’s insurer, a public adjuster and State Farm, her homeowners insurance company, as defendants. The AI problem surfaced two years later, inside the motions that were supposed to shape what a jury would eventually hear.
What the plaintiff’s team found
Eric Khodadian, representing Meni-Siliga, set out the discovery in a court filing: the defence motions relied on cases, quotes and holdings that could not be located anywhere because none of them were real. Pulling and reading cited authority is basic litigation hygiene, and it is the step that has caught nearly every one of these incidents since the first ones surfaced in 2023.
Kenneth Katel of Musick, Peeler and Garrett apologised to the court and to the plaintiff. By his account, a co-counsel used AI to help prepare some of the filings and he did not know it had been used. He did not hide behind that. “As lead trial counsel I accept full responsibility for our filings,” he wrote.
The supervision gap
What makes this one notable is not the hallucination. Fabricated citations have turned up in filings repeatedly since 2023, and judges have accumulated a small body of sanctions decisions in response. What stands out is where it happened: a defence firm acting for one of the largest insurers in the United States, on a case heading to trial, using software built specifically for legal research.
Purpose-built legal tools were supposed to close this failure mode. Retrieval-based research products ground their answers in actual case databases rather than a model’s memory, which is precisely the safeguard general-purpose chatbots lack. That architecture lowers the error rate. It does not remove the need for a human to open the case and read it, and this incident suggests verification is the step that keeps getting skipped when deadlines compress.
A mechanical fix has been available the whole time. Citators such as Shepard’s and KeyCite have let lawyers confirm that a case exists and remains good law for decades, and running a brief through one takes minutes. What generative drafting changed is the volume of citations arriving without a human having read the underlying opinion, which turns a routine check into the only thing standing between a draft and a sanctionable filing.
Pressure from the client side
The pressure driving that shortcut is measurable. Thomson Reuters found that 41 percent of law firms and 47 percent of corporate legal departments now report their legal teams using generative AI, up from 28 percent and 23 percent a year earlier. Thirty-eight percent of law firm professionals said they feel significant or some financial pressure to move faster on AI. And 32 percent of in-house legal professionals said they are already reconsidering relationships with firms that cannot show clear AI-enabled value within twelve months.
Read those three numbers together and the shape of the problem is hard to miss. Clients are demanding AI-driven efficiency, firms are adopting quickly to keep the work, and the verification burden that adoption creates lands on the same associates who were already short of hours.
Regulators are working to a different clock. In Europe, the EU AI Act’s high-risk obligations took effect on 2 August 2026, bringing mandatory logging, documentation and human oversight requirements to a defined set of systems, with fines reaching €15 million or 3 percent of global annual turnover. American courts have no equivalent framework. They have judges, standing orders on AI disclosure, and their own sanctions powers.
Judges have started closing the gap themselves. Standing orders requiring lawyers to disclose AI use in filings, or to certify that every citation has been verified by a human, have spread across US federal and state courts since 2023 without any central rule change. The result is a patchwork: the obligation on a lawyer now depends heavily on which courtroom the case lands in.
Sanctions in this matter have not been decided. The more interesting question is what happens to firm-level policy across insurance defence practices, where high volume and thin margins make AI-assisted drafting genuinely attractive. A rule requiring associates to verify every citation is easy to write. Enforcing it, when the entire appeal of the tool was saving the hour that verification takes, is the part nobody has solved.
For more coverage of AI and the legal profession, visit Mylistingo.
