Courts Draw the Line on AI: July 2026 Industry Roundup
An Indiana appellate court flags apparent AI errors in a court reporter's transcript, Arizona rules that citing fake AI cases is sanctionable regardless of intent, and an Ohio suit tests whether AI transcripts are discoverable like email. The thread this edition: courts are putting the accountability on the person using the tool, not the tool itself.
Alicia Moffatt
· 4 min read
This edition, courts and regulators stopped debating AI in the abstract and started drawing lines. Appellate judges flagged apparent AI errors in a trial transcript, a court made fabricated AI citations sanctionable no matter the filer's intent, and a discovery fight is testing whether AI transcripts are evidence like any other email. TheRecordXchange® reads the coverage so you can stay oriented. Below are the stories worth your time, ranked by relevance to the courts we serve and by the authority of the source.
The three to read first
Reason — The Volokh Conspiracy
Court Notes Apparent AI-Generated Errors in Court Reporter's Transcript
In Williams v. State, the Indiana Court of Appeals flagged what it described as apparent generative-AI involvement in a trial transcript riddled with errors: meaning-altering typos, wrong names for witnesses and attorneys, and misattributed motions, objections, and even a closing argument. The court said the mistakes complicated but did not substantially impede its review, and it reminded the reporter that professionals using AI must proofread and verify the output. For a court that depends on the transcript being a true and accurate record of the proceeding, it is a pointed reminder that accuracy is the reporter's responsibility, not the tool's.
Arizona Mirror
AZ Court of Appeals: Citing Fake AI-Generated Cases Can Get You Sanctioned, No Matter Your Intent
The Arizona Court of Appeals held that citing fake, AI-generated cases can draw sanctions regardless of the filer's intent, so a good-faith mistake is no defense once fabricated citations reach the court. The ruling puts the burden squarely on the person filing to verify every authority, whether or not AI was involved. It adds to a growing line of decisions that treat unverified AI output in court filings as a professional-responsibility failure rather than an honest error.
RealClearPolicy
The Court Case That Could Change AI at Work
An Ohio dispute between a fired executive team and the investor who dismissed them is testing whether AI-generated transcripts of workplace conversations are discoverable evidence, much like email. The question is whether recordings that automatically save to company systems carry any privilege or expectation of privacy once litigation begins. However it resolves, the case signals that AI transcripts are becoming part of the discoverable record that courts and litigants will have to account for.
More from this edition
Axios
A Northwestern survey finds most federal judges have now tried AI in their work, though few reach for it daily, a picture of a bench that is curious but still cautious.
Stanford HAI
Stanford's Institute for Human-Centered AI examines legal AI's legibility problem: the difficulty of understanding, explaining, and trusting how these systems reach their outputs.
Thomson Reuters Institute
A look at how reverse mentorship, with junior staff coaching senior judges and lawyers, is helping modernize the bench without waiting for a generational turnover.
BW Legal World
India publishes draft rules for AI in its courts while the UK's Garfield AI draws the line regulators keep returning to: AI can assist judges, but it cannot adjudicate.
The Hill
Illinois becomes the first state to mandate third-party audits of frontier AI labs under its new AI Safety Measures Act, a template other states may follow.
Why we publish this
TRX serves the court community, and part of that service is helping you stay oriented as courts and regulators decide how AI belongs in the record. We read widely, rank by relevance and source authority, and pass along what is worth your attention. This roundup is curated and published when enough worthwhile stories accumulate. If you have a story we should consider for the next edition, send it our way.