The Reasonable Inquiry — October 5, 2026
Subscribe to The Reasonable Inquiry
October 5, 2026 · David Bartlett
Courts are done asking whether lawyers use AI. They're asking who checked.
AI in the Courts
2,149 court decisions now involve AI-hallucinated citations. That count is from one researcher's database, and it grows every week. (AI Hallucination Cases, checked October 5)
The courts have stopped debating whether lawyers use AI. The rules now assume you do. They ask one question: who checked?
Here is where the answer stands this month.
Florida made verification a signature. Since June 15, every filing in Florida state court certifies that "the legal authorities identified exist and are accurately cited." Not that you tried. That they exist. (Florida Bar News)
The Tenth Circuit wants a human on the hook. On September 18 it proposed amending Rule 46.5 "to address the increased use of generative artificial intelligence by both lawyers and pro se litigants." Filers who used AI would certify that a person verified the authorities, quotes and analysis. Comments close October 18; the rule would take effect January 1. (Tenth Circuit)
Trial judges aren't waiting for appellate rules. Two Central District of California orders on September 4 put it this way: "Failure to submit a declaration attesting to the use or non-use of AI may result in the Court striking the filing." Not a sanction. A strike. (EDRM)
The federal evidence rule on AI stalled. Proposed Rule 707, meant to govern machine-generated evidence, went out for comment, came back, and in May the Advisory Committee wrote: "The Committee does not recommend action on the proposed Rule 707 at this time." It wants AI experts to vet it first. The earliest it could now take effect is years out. (Advisory Committee report)
Arizona's Supreme Court refused to ban judges from using AI. On August 27 it rejected a petition for a moratorium and kept its controlled testing, with one line that should be pinned above every bench and every desk: "Decision-making is a human function and a core judicial responsibility that cannot be delegated to AI or anyone else." (Arizona Supreme Court)
Read those five together and the pattern is plain. The rules don't restrict the tool. They assign the blame. Every one of them ends at a human name.
Monday takeaway: if your firm has no written rule on who verifies AI output before it's filed, you have one anyway. It's whoever signs.
AI Legal Stories...
The judge who asked ChatGPT
An Oklahoma judge signed a custody order citing two cases that don't exist. He wrote it with ChatGPT.
Associate District Judge Lawrence Wheeler of Stephens County told a state investigator he used the chatbot for research. The order went out with the fake citations in it. The Attorney General reviewed the investigation and declined to bring criminal charges. Wheeler recused himself from the case. Discipline, if any, now sits with the state's judicial bodies. (Volokh Conspiracy / Reuters; News on 6)
For two years the story was lawyers filing fake cases and judges catching them. This is the first widely reported case of the catching going the other way. The difference matters: a lawyer's hallucination costs a client. A judge's becomes an order.
The dead man's testimony
An Arizona judge let a dead man speak at his killer's sentencing. On September 30, the Court of Appeals said he shouldn't have.
The conviction stands. The sentence doesn't.
In 2021, Gabriel Horcasitas shot and killed Army veteran Christopher Pelkey in a Chandler road-rage confrontation. At sentencing in May 2025, Pelkey's family played a video of Pelkey addressing the court. The video was AI. The words were his sister's. The judge said he "loved" it, then imposed 10½ years for manslaughter.
The State argued the video was reliable because it reflected what the sister believed her brother would have said. Division One saw it differently. It called the video "the imaginings of the victim's sister" and held that the judge's "consideration of the AI video so prejudiced Horcasitas as to render the sentencing procedure fundamentally unfair."
Arizona's constitution gives victims the right to be heard. This opinion marks where that right ends: with people speaking for themselves. A model speaking for the dead isn't a victim statement. It's a script with a face on it. (State v. Horcasitas, No. 1 CA-CR 25-0191, opinion)
The government's case that never existed
A federal prosecutor cited Taylor v. Hott, 724 F. App'x 387, in an immigration bond case. There is no such case.
Chief Judge Hala Y. Jarbou of the Western District of Michigan found the citation was likely produced by generative AI. She let it go without sanctions this once, then set the bar for next time: "any attorney who uses AI must scrupulously review its work product to ensure that the cited cases exist and that the citations accurately and fairly represent the underlying case law." (Daghra v. Hinkley, No. 1:26-cv-01429; Bloomberg Law)
The man in that case was in detention. The brief arguing to keep him there rested on a precedent that doesn't exist. Nobody at the Department of Justice checked. The judge did.
Monday takeaway: every citation gets opened before it gets filed. Not searched. Opened.
New Law
On September 30, California became the first state to put a lawyer's duty to check AI into a statute. Not a bar opinion. Not a standing order. A law.
SB 574 passed both houses without a single no vote and takes effect January 1, 2027. The operative line: "An attorney shall not delegate the practice of law to generative artificial intelligence." Lawyers must "take reasonable steps to verify the accuracy of AI outputs, including, but not limited to, the accuracy of all case and statutory citations," and must "correct any erroneous or hallucinated output." Confidential information stays out of open AI tools. AI use gets disclosed to the court. (Holland & Knight; Bloomberg Law)
Notice what the statute does not do. It doesn't ban the tool. It doesn't license it. It names the person responsible for what the tool says. That is the same move the courts made above, now written by a legislature. Expect copies in other states within a year.
Two more from the same week, both on the same theme: when a machine decides, a human answers for it.
- California SB 947, the "No Robo Bosses Act" (signed September 30, effective July 1, 2027): before an automated system can get a worker disciplined or fired, an employer needs "a human reviewer to independently corroborate the ADS output." The governor vetoed last year's version. This one he signed. (Ogletree)
- Connecticut's CART Act (Public Act 26-15): first duties took effect October 1, including whistleblower protection for people inside frontier AI labs and watermarking for large AI content providers. Employers must tell workers when AI drives a hiring or firing decision starting October 2027. Its sponsor's own description: "This is the floor." (CT Mirror; Ropes & Gray)
And one retreat: Colorado repealed its broad 2024 AI Act in May and replaced it with a narrower notice-and-disclosure law effective January 1, 2027. The ambitious version never took effect. (Seyfarth)
The legislatures are converging on one sentence, and it isn't about AI at all: somebody has to be accountable, and it can't be the software.
Still on the Books
In Minnesota, greasing a pig and letting people chase it is a crime. It has been since 1971. It still is.
Minnesota Statutes § 343.36, in full:
"No person shall operate, run or participate in a contest, game, or other like activity, in which a pig, greased, oiled or otherwise, is released and wherein the object is the capture of the pig, or in which a chicken or turkey is released or thrown into the air and wherein the object is the capture of the chicken or turkey. Any violation of this section is a misdemeanor."
That's the 2025 edition, on the state's own site. (Minnesota Revisor of Statutes)
The reflex is to laugh. Hold it a second, because the law has three things to say to a practitioner, and none of them is funny.
First: most of these are fake, and this one isn't. The internet runs on lists of "dumb laws," and the lists copy each other, not the code. Wikipedia's own entry on the genre concedes that "a large number of hoax or exaggerated dumb laws are circulated on the internet and in the print media." The donkey-in-a-bathtub law, the ice-cream-in-your-pocket law: nobody can produce the section. This one has a section. The difference between the two is the difference between a rumor and a charge.
Second: nobody enforces it, and that changes nothing. Desuetude is not a defense in an American court. A statute dies one way: a legislature repeals it, through the same bill, committee, floor vote and signature it took to pass. Until then, it waits. Wisconsin's doctors learned what waiting means on June 24, 2022. An 1849 statute, Wis. Stat. § 940.04, that no one had enforced in fifty years was suddenly the operative law of the state the afternoon Dobbs came down. Providers stopped. They stayed stopped for three years, until the Wisconsin Supreme Court held, 4-3, on July 2, 2025, that later statutes had impliedly repealed it. (State Court Report) Three years of a dead law running a state.
Third: some of them are enforced every week. In Colorado it is illegal to sell a car on Sunday. Not a legend. Colo. Rev. Stat. § 12-6-302, now recodified in Title 44: no person "shall keep open, operate, or assist in keeping open or operating any place or premises" for selling a motor vehicle "on the first day of the week commonly called Sunday." Misdemeanor, up to $1,000 and six months, and the dealer's license on the line. A state review board recommended repeal in 2017. The dealers' own association lobbied to keep it, and won. (Justia, C.R.S. § 12-6-302; KDVR; CPR)
So the greased pig isn't a joke about Minnesota. It's a reminder of how the code works. A statute means what its words say, whether or not anyone has read them lately. A prosecutor with a grudge, a court that changes its mind, or a trade group that likes things as they are can bring any of them back to life without passing anything.
Monday takeaway: when a client asks "is that still a law?", the answer is never "nobody enforces that." The answer is the section number, read today.
Every statute on ClearStatute shows its official text and the date it was last checked against azleg.gov. Arizona only, for now.
Next Issue...
Is the statute you cited last week still the law?
We ran all 22,619 Arizona statutes against the Legislature's own text. Something didn't add up. How many of the free copies the internet runs on are wrong? How long have they been wrong? And whose brief cited one last month?
The Reasonable Inquiry is published by Legal Search Hub LLC, maker of Cite11 and ClearStatute. It reports on the law; it is not legal advice.
Sources
- Damien Charlotin, AI Hallucination Cases database
- Florida Bar News: Supreme Court amends rules to address AI use in court filings
- Tenth Circuit: public comment period for 2027 rules
- EDRM / Michael Berman on standing orders requiring AI disclosure
- Advisory Committee on Evidence Rules, May 2026 report
- Arizona Supreme Court release on judicial AI use
- Volokh Conspiracy: Oklahoma judge's ruling contained AI hallucinations
- News on 6: Oklahoma judge recuses
- State v. Horcasitas, 1 CA-CR 25-0191, opinion
- Bloomberg Law: DOJ called out by judge for fake case citation
- Holland & Knight on California SB 574
- Bloomberg Law: Newsom signs first-of-its-kind bill on lawyer AI use
- Ogletree on SB 947 and companion bills
- CT Mirror on Connecticut laws taking effect October 1
- Ropes & Gray on the Connecticut CART Act
- Seyfarth on Colorado's replacement AI law
- Minnesota Statutes § 343.36
- State Court Report: Wisconsin Supreme Court rules 176-year-old law does not ban abortion
- Justia: Colo. Rev. Stat. § 12-6-302, Sunday closing
- KDVR: Why can't you buy a car on Sundays in Colorado?
- CPR: a state agency recommends changing the Sunday car-sales law
- Wikipedia: Strange laws (on hoax lists)