AI Court Rules: How judges are policing AI in filings
When a brief carries your signature, the citation is yours. We track the orders, sanctions, and standing rules that decide what an AI-assisted filing must prove before it leaves your office.
A topic page in AI Regulation News, our source-backed tracker of the rules that change how professionals use AI at work.
The fastest moving corner of AI regulation is not a statute. It is the bench. Across federal and state courts, judges are deciding case by case what an attorney owes when a brief is drafted with help from a generative tool, and the answers are arriving as sanctions orders, standing rules, and revoked admissions rather than as legislation. For a practicing lawyer, this is the part of the law that changes your Monday before any agency finishes a rulemaking. A judge in your district can set the rule for your next filing this week.
The pattern across these rulings is consistent, and it is narrower than the headlines suggest. Courts are not punishing the use of AI. They are punishing the failure to verify. The recurring fact pattern is a citation to a case that does not exist, or a real case cited for a holding it never reached, signed and filed without anyone confirming it against a primary source. When that surfaces, the court reaches for Rule 11 or its state equivalent and treats the missing verification step, not the tool, as the sanctionable conduct. The duty attaches to the human who signs, and it does not transfer to the model or to co-counsel.
What is at stake has climbed. Early hallucination cases produced a fine and a public scolding, a cost most firms could absorb. The newer orders reach further. We have seen pro hac vice admissions revoked, two-year bars from a district, mandatory AI ethics education, and separate candor findings stacked on top when a lawyer was less than straight about how the work was done. A fine is a line item. Losing the right to appear in a forum is a hole in a practice. Some courts are also moving from after-the-fact punishment toward front-end rules, issuing standing orders that require disclosure of AI use or certification that every authority was checked.
The workflow consequence is the same whether your court has spoken yet or not. The defensible position is not avoiding AI and it is not trusting it. It is a short, documented routine that runs on every filing: pull each authority from a primary source, confirm the holding and that it is still good law, name the verifier of record, and log the check before the signer's final pass. Treat the loud cases below as a preview of the standard arriving in your jurisdiction, and build the routine before a judge builds it for you. This page tracks the orders that move that line. It is professional workflow interpretation, not legal advice.
The briefings we track in this lane
A Mississippi Judge Barred Two Lawyers for Two Years Over AI Citations
Judge Aycock revoked two pro hac vice admissions and imposed a two-year bar after both sides filed AI-fabricated citations in Withers v. City of Aberdeen.
Read the briefing Court rulingThe Ninth Circuit AI Sanctions Ruling and What Your Filings Now Require
An appellate signal on AI-fabricated authority, and the verification standard it sets for every brief filed in the circuit.
Read the briefing Court ruleFlorida Rule 2.515 and the June AI Filing Sanctions
Florida moved the certification duty into a filing rule. What Rule 2.515 asks attorneys to confirm before signing an AI-assisted document.
Read the briefing LitigationThe Anthropic Settlement and Where the Fair Use Line Falls
A federal ruling held that training on lawfully acquired books can be fair use while hoarding pirated copies is not. What it means for what you feed AI.
Read the briefing Bar guidanceThe California Bar Mapped Agentic AI to Your Supervision Duty
California's first agentic AI guidance maps autonomous workflows to Rules 1.1, 1.6, and 5.3, with a supervision protocol for tools that act on their own.
Read the briefingBrowse the full tracker for every rule, ruling, and agency action we follow across professions.
Keep going
- Court Puts Army AI Evaluations in TRAX Protest RecordSealed 3 September, reissued 22 September 2026: the TRAX ruling put Army FAST TRACK AI evaluations in the record but denied the protest on the merits.
- Tenth Circuit Proposes AI Filing CertificationProposed Rule 46.5(B)(4): human verification of every cited authority in AI-assisted filings; binds nobody until adopted.
- Fed. Circuit Revives VA AI Tool Whistleblower CasePrecedential: disclosures about a flawed VA AI claims tool were nonfrivolous gross-mismanagement allegations.
- Idaho Supreme Court: AI No Excuse for Bad BriefsAI use does not excuse briefing errors by lawyers or pro se parties; the Court imposed no sanction.
- Montana Deepfake Law Enjoined for Two PlaintiffsThe three named officials can no longer act under the Act against the two plaintiffs.
- Russia Supreme Court Order 59-P Sets AI Courts PlanA dated instruction from the Chairman of the Supreme Court now exists approving a court-system-wide Concept for AI in judicial proceedings, with a calendar running from the third quarter of 2026 to the second q
- EAT Lists Minimum AI Checks for Appeal DocumentsAn EAT judgment now records, on facts where a litigant in person filed a 300-page, almost 132,000-word skeleton created with ChatGPT, that the author was responsible for checking it against the Practice Directi
- Fla. 4th DCA Show-Cause Order Over AI Slop FilingsA Florida district court of appeal has issued a show-cause order against a licensed attorney over filings it suspects contain AI-generated components, on the ground that the arguments are frivolous and contrary
- Ninth Circuit Affirms DMCA Dismissal in Doe v. GitHubA published Ninth Circuit ruling on whether, on this complaint, an AI code generator's output is a copy from which CMI was removed or altered.
- NM Court Holds Lawyer in Direct ContemptA state supreme court used direct contempt, grounded in its constitutional power of superintending control, rather than litigation sanctions, and the fabricated material included witness testimony and a factual
- Arizona Supreme Court No. R-26-0020, 27 August 2026Arizona Supreme Court Order No. R-26-0020, 27 August 2026. Denies a petition to adopt a new Rule 135 without stating a reason, states that no tool can relieve judicial officers of full accountability for their decisions, and requires an AI steering committee report by 1 March 2027.
- F.J.L v T.G.O (2025/220239) [2026] ZAGPJHC 875, Gauteng Division, Johannesburg, 31 July 2026A delivered and reportable High Court judgment, binding on the parties. The referral is not a finding of misconduct: the concurring judge records expressly that what he sets out are observat
- Agravo de Instrumento no 5013085-69.2025.4.03.0000, TRF-3, Quarta Turma, agravo internoBinding between the parties. The Fourth Panel unanimously gave partial relief on internal appeal: the fine stands at 5 percent of the value of the principal claim, but it is now owed by the
- Landgericht Muenchen I, 42. Zivilkammer, Urteil vom 31.07.2026, Az. 42 O 763/25 (GEMA v Suno)A first instance judgment, and expressly not final. The court's own press release states that the judgment is not legally binding, which means an appeal remains open.
- Decision of the chairman of the Central Elections Committee for the 26th Knesset in TBK 73/26 and TBK 74/26Binding on the parties before the committee. This is a decision on two petitions under section 17 of the Elections (Means of Propaganda) Law, and it grants removal plus costs, not guidance.
- Practice Direction CA16, Practice Direction on the responsible use of Generative Artificial Intelligence in CoBinding on practice before the Court of Appeal. Ms Justice Caroline Costello issued it as President under the general authority of the office, the Court's inherent power to regulate its own
- SRA Warning Notice on the misuse of AI by solicitorsThe SRA says AI has no separate legal personality and that reliance on an AI output is not a suitable defence for false citations put before a court.
- Indiana Supreme Court Order 26S-MS-8, amending Trial Rule 11Since 1 August 2026 an Indiana signature certifies that any cited legal authority is authentic, and the rule expressly covers AI-assisted filings.
- Read how a Mississippi judge barred two lawyers for two years over AI citations
Recent court and tribunal decisions
- Texas Court Weighed Striking a Brief Over Fake CitesSAISD v. Becerra: four of eight cited opinions appeared to be AI hallucinations, and the court weighed striking the brief under Rule 38.9.
- Delaware: Fake Quotes Are an AI HallmarkPalumbo v. Palumbo: Delaware Chancery treats fictitious quotations as a hallmark of AI use and calls abuse of AI in filings sanctionable conduct.
- Pa. Court: Fabricated Citations Waive the ClaimKaspryak v. Stadarskyy, 2026 PA Super 185: hallucinated citations led to waiver of the claims for failure to cite pertinent authority, not a sanction.
- Tribunal Rejects AI Claim Against Expert ReportIn LR (Male VOTs) Albania CG the Upper Tribunal found no proper foundation for a cross-examination allegation that an expert report was AI-generated.
- Virginia Court Orders AI Training After Fake QuotesVirginia Court of Appeals found 25 phantom quotes in a brief and ordered a CLE course on AI instead of a fine under Code 8.01-271.1.
- Court Doubts ChatGPT-Polished Witness EvidenceGodwin v Godwin: the court could not be sure witness statements were in the makers' own words after ChatGPT use, and treated the evidence cautiously.
- Judge: AI Invented an Insolvency Rule in Firm LetterICC Judge Mullen found a Pinsent Masons letter quoted a non-existent Insolvency Rule produced with AI, and ordered the judgment sent to the SRA.
- Upper Tribunal: A Recommender Has No Artistic PurposeUpper Tribunal held TikTok's recommender did not process data for the special purposes, dismissing its appeal against the ICO penalty.
- High Court Rejects AI Excuse for Faked Court ExhibitLinden J referred Amar Lodhia to the Attorney General over falsified evidence, rejecting his claim that an AI tool generated a fabricated exhibit.
- 7th Circuit: AI Images With No Real Child, As AppliedIn Anderegg the Seventh Circuit struck one possession count as applied. The government's concession that no real child was depicted drove the result.
- Brazil Lets Courts Use AI to Hunt Debtor AssetsProvimento 255 lets Brazilian courts use AI for asset research and drafts in execution, but bars it from decisional acts. In force around 19 September.