A Stenographer Submitted AI-Generated Errors in Official Court Transcript, Judge Says

Court reporters have long been the one job in the legal system nobody worried about when it came to AI slop. Lawyers messing up citations with a chatbot, sure, that's old news by now. But a stenographer's transcript getting flagged for AI fingerprints? That's new territory, and it happened in a Florida appeals case decided July 23.

Judge Paul Felix, reviewing an appeal from a man convicted in a fatal overdose drug sale, found a transcript riddled with attribution errors that read like a language model guessing at who said what. A motion presumably filed by the State got credited to the trial court. An objection from the defendant, Williams, got pinned on the bailiff. The State's closing argument was, again, attributed to the judge. Felix noted in a footnote that these mistakes didn't derail his review of the appeal, but he clearly wasn't thrilled, writing that the errors bore the hallmarks of generative AI having a hand in producing the transcript.

Attorney Rob Freund spotted the footnote and posted it on X, and it's easy to see why it caught fire. For a few years now, judges have been publicly shaming lawyers for citing AI-invented case law, including a particularly rough 20-minute dressing-down in May from New York's Appellate Division, where judges called one filing situation striking, disappointing, and saddening, all in the same breath. Lawyers usually blame a paralegal, an illness, deadline pressure — the excuses are practically their own genre at this point.

What makes this different is the stakes. A lawyer's brief with a fake citation is embarrassing and sanctionable, but a transcript is supposed to be the ground truth of what actually happened in a courtroom. If AI tools are quietly filling gaps in stenography with plausible-sounding guesses instead of flagging uncertainty, that's not a citation error, that's a corrupted record of a trial. Stenographers have always argued their value is in precision under pressure, catching mumbled words, interruptions, and overlapping speech that a machine would rather smooth over than question. Felix's footnote is the first public sign that when AI does the smoothing instead of a human, the record itself can't be trusted the way courts assume it can.