Search significant federal and published state decisions that materially address artificial intelligence, automated decision systems, or algorithmic evidence. Each entry identifies the holding, procedural posture, and a reviewable opinion source.
24 of 24 curated opinions
Last editorial review: September 15, 2026
FederalAI outputs & liability
James Keene v. Google LLC
No. 1:25-cv-11431, Dkt. 29 (N.D. Ill. Sept. 14, 2026) · U.S. District Court, Northern District of Illinois
Holding
At the pleading stage, the court held that AI Overviews stating that Keene was serving life sentences for multiple convictions or murders plausibly conveyed actionable factual assertions; neither their AI origin, linked sources, nor a disputed error disclaimer made them nonactionable opinion. Repeated notice followed by similar alleged publications plausibly supported actual malice. The court dismissed a third Overview as substantially true and a fourth for failure to plead third-party publication, both without prejudice. It did not adjudicate falsity, actual malice, or liability on the merits.
Why it matters
A significant AI-output-liability ruling that treats summaries embedded in ordinary Google Search as potentially factual, distinguishes deliberately prompted chatbot output, and permits core defamation claims arising from allegedly hallucinated AI search summaries to proceed.
Posture
Rule 12(b)(6) motion in a removed defamation-per-se action concerning four Google AI Overviews
TOV Realty, LLC v. Suarez / Kosel Equity, LLC v. MacGregor
355 Conn. 902–909, Nos. SC 21183 & SC 21184 (order July 31, 2026; officially released Aug. 4, 2026) · Connecticut Supreme Court
Holding
The court held that counsel violated Connecticut Rule of Professional Conduct 1.1 by filing seven erroneous or unverified citations introduced when ChatGPT edited previously verified drafts. Finding no intent to deceive, it ordered six additional CLE hours, including three concerning generative AI; required counsel and the firm each to donate $1,000 to the Connecticut Bar Institute; and imposed compliance-reporting and reciprocal-discipline obligations. The court did not decide whether other professional-conduct rules were violated.
Why it matters
The Connecticut Supreme Court's first direct ruling on generative-AI hallucinations establishes that lawyers must reverify citations after AI-assisted editing—not merely before it. The published order also identifies confidentiality, attorney-client privilege, work-product, supervision, and firm-policy concerns as aspects of competent AI use.
Posture
Published per curiam sanctions order following judgments in two public-interest appeals and a sua sponte show-cause hearing
No. CAAP-24-0000826 (Haw. Ct. App. Aug. 31, 2026) · Hawaiʻi Intermediate Court of Appeals
Holding
The published court held that arguments relying on nonexistent authorities may be treated as waived under Hawaiʻi Rules of Appellate Procedure 28(b)(7), and that each false citation can support sanctions under Rules 30 and 51. It declined to import civil Rule 11 into the direct criminal appeal and imposed no sanction, instead reaching the remaining discernible issues and affirming the conviction. The court discussed the rise of AI-associated fake citations but did not find that Presti used AI.
Why it matters
A precedential state appellate decision establishing an operative waiver rule for arguments built on fabricated authorities while distinguishing civil sanctions doctrine from the constitutional setting of a self-represented criminal appeal.
Posture
Self-represented criminal defendant's direct appeal from a second-degree unlawful-imprisonment conviction
LPC Commercial Services, LLC v. Ar-Razzaaq Properties, LLC
Record No. 1407-25-4 (Va. Ct. App. Aug. 25, 2026) · Court of Appeals of Virginia
Holding
After finding 25 invented or materially inaccurate quotations in the appellee's brief, the published court inferred that unchecked computer-assisted research caused the errors and applied Virginia Code § 8.01-271.1. It ordered counsel to complete a CLE course concerning AI or appellate practice within six months, while declining a monetary consequence and affirming the underlying judgment.
Why it matters
A precedential state appellate decision imposing a concrete professional-responsibility consequence for unchecked computer-assisted legal research and reaffirming that the lawyer signing a filing remains responsible for its accuracy.
Posture
Appeal from a breach-of-contract judgment following a Fairfax County Circuit Court bench trial
Civil No. 26-3425 (DWF/DTS), Dkt. 54 (D. Minn. Sept. 4, 2026) · U.S. District Court, District of Minnesota
Holding
The court denied preliminary relief because xAI delayed seeking an injunction, did not demonstrate likely irreparable harm, and the equities and public interest favored the State. It expressly deferred standing, scrutiny, and the First Amendment merits to later proceedings.
Why it matters
An early federal ruling on emergency relief from a state law aimed directly at providers of AI nudification tools, illustrating that delay and weak evidence of irreparable harm may defeat preliminary relief without resolving constitutionality.
Posture
Motion for preliminary injunction against Minnesota's AI-nudification law
No. 24-CV-1099 (D.C. Sept. 3, 2026) · District of Columbia Court of Appeals
Holding
The court struck the appellee's brief under D.C. Appellate Rule 28 after counsel confirmed that four cited cases did not exist and admitted using Google generative AI without independently verifying the results. It referred the matter to disciplinary counsel without finding knowing, reckless, or bad-faith conduct.
Why it matters
A published order from the District's highest court applying appellate briefing requirements directly to fabricated generative-AI authorities while preserving the distinction between striking a defective filing and imposing attorney discipline.
Posture
Published sua sponte order during a judicial-foreclosure appeal
No. 3:26-cv-01996-RFL, Dkt. 250 (N.D. Cal. Aug. 27, 2026) · U.S. District Court, Northern District of California
Holding
The court granted Anthropic partial summary judgment on First Amendment retaliation and Fifth Amendment due-process claims and held that the challenged Department of War directive and supply-chain-risk designation exceeded statutory authority and were procedurally defective and arbitrary or capricious. The government prevailed on other defendants and theories, including the ultra vires separation-of-powers claim.
Why it matters
A major merits judgment applying constitutional and Administrative Procedure Act limits to federal procurement sanctions imposed on an AI developer because of its public AI-safety position and restrictions involving autonomous weapons and mass surveillance.
Posture
Cross-motions for summary judgment challenging federal procurement and supply-chain measures
Disruptive Resources, LLC v. Ballistic Barrier Products Inc.
No. 1:24-cv-00321-JCG, D.I. 115 (D. Del. Aug. 20, 2026) · U.S. District Court, District of Delaware
Holding
The court declined to sanction counsel after counsel promptly disclosed generative-AI errors, accepted responsibility, and corrected the affected briefing. The court did not decide that Rule 11 had not been violated and warned that future AI mistakes in the case may result in sanctions.
Why it matters
A concrete federal resolution of an AI-drafting sanctions proceeding showing how candor and prompt correction informed the court's discretionary decision, while preserving lawyers' duty to thoroughly verify AI-assisted filings.
Posture
Rule 11 show-cause proceeding concerning an AI-drafted claim-construction brief
No. 4:25-cv-06328-JST, ECF 74, 2026 WL 21791 (N.D. Cal. Jan. 5, 2026) · U.S. District Court, Northern District of California
Holding
The court dismissed the complaint with prejudice. It held that the plaintiff failed to plead reasonable measures to preserve trade-secret secrecy because she alleged that she developed the claimed protocols and frameworks through ChatGPT, thereby voluntarily disclosing them to OpenAI without an alleged duty of confidentiality. Contractual language assigning ownership of outputs did not establish secrecy.
Why it matters
A material ruling on whether entering claimed proprietary methods into a generative-AI service preserves trade-secret protection. The Ninth Circuit later dismissed the appeal for failure to prosecute, not on the merits; that dismissal became the mandate on April 27, 2026.
Posture
Rule 12(b)(6) motion to dismiss a pro se first amended complaint
Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc.
No. 20-613 (D. Del. Feb. 11, 2025) · U.S. District Court, District of Delaware
Holding
ROSS's use of copied Westlaw headnotes to build a competing legal-research system was not fair use; selected headnotes and the Key Number System received differing copyright treatment.
Why it matters
The first major U.S. merits ruling rejecting an AI-training fair-use defense on a developed record, with emphasis on commercial substitution and the nature of the copied material.
No. 3:24-cv-05417-WHA (N.D. Cal. June 23, 2025) · U.S. District Court, Northern District of California
Holding
Training language models on authors' books was fair use on the record presented, as was converting lawfully purchased print books to digital form; retaining pirated copies in a central library was not justified by the training purpose.
Why it matters
A foundational ruling separating the purpose of model training from the legality of acquiring and retaining the training corpus.
No. 3:23-cv-03417-VC (N.D. Cal. June 25, 2025) · U.S. District Court, Northern District of California
Holding
Meta prevailed on fair use because the authors did not develop evidence sufficient to show relevant market harm, while the court cautioned that the ruling did not establish that AI training is categorically lawful.
Why it matters
Places market substitution at the center of generative-AI fair-use litigation and illustrates how the evidentiary record can control the result.
No. 3:23-cv-00201-WHO (N.D. Cal. Aug. 12, 2024) · U.S. District Court, Northern District of California
Holding
The court allowed core direct-infringement and related theories concerning Stable Diffusion training and model operation to proceed while dismissing or narrowing other claims.
Why it matters
A leading pleading-stage decision on whether model training, model copies, and generated outputs can support copyright and Lanham Act theories.
No. 3:23-cv-03223-AMO (N.D. Cal. Feb. 12, 2024) · U.S. District Court, Northern District of California
Holding
The court dismissed vicarious-infringement, DMCA, negligence, unjust-enrichment, and unfair-competition theories with varying leave to amend, while direct-infringement claims remained outside that motion.
Why it matters
An early, influential test of generative-AI copyright pleading theories and the distinction between alleged training copies and allegedly infringing outputs.
No. 3:23-cv-00770-RFL (N.D. Cal. July 12, 2024) · U.S. District Court, Northern District of California
Holding
An AI hiring-tool vendor could plausibly act as an employer's agent when customers delegate traditional hiring functions to its automated screening tools, allowing portions of the discrimination case to proceed.
Why it matters
The leading U.S. decision on potential direct liability of an automated employment-decision vendor under federal anti-discrimination law.
114 Cal. App. 5th 421 (2025) · California Court of Appeal, Second District
Holding
The court imposed sanctions after counsel filed briefs containing fabricated and mischaracterized authorities generated with AI and failed to verify them.
Why it matters
California's first published appellate warning focused on hallucinated legal authority and counsel's nondelegable duty to read and verify cited cases.
No. E083744 (Cal. Ct. App. Nov. 17, 2025) · California Court of Appeal, Fourth District
Holding
The court found that multiple spurious citations bore the hallmarks of AI hallucinations, rejected counsel's clerical-error explanation, and imposed consequences for the unsupported filing.
Why it matters
Extends California's published AI-citation jurisprudence and demonstrates that implausible explanations can compound the underlying verification failure.
No. D084581 (Cal. Ct. App. Oct. 2, 2025) · California Court of Appeal, Fourth District
Holding
The court addressed counsel's admitted failure to verify AI-generated citations and treated the false presentation of authority as a breach of professional obligations.
Why it matters
Applies AI-verification principles in criminal practice, where inaccurate authority can directly affect liberty interests.
Posture
Published criminal appeal addressing attorney misconduct
No. D085584 (Cal. Ct. App. Mar. 5, 2026) · California Court of Appeal, Fourth District
Holding
The court emphasized that both lawyers and courts share responsibility for protecting the legal system from fabricated law and dismissed an appeal infected by invented authority.
Why it matters
A current California statement that institutional safeguards against hallucinated citations extend beyond individual counsel.
A sentencing court could consider a proprietary algorithmic risk score with warnings and limitations, but could not use it to determine incarceration or sentence length as the decisive factor.
Why it matters
The foundational state high-court decision on due process, transparency, bias, and human responsibility when government decisions incorporate algorithmic risk assessment.
466 N.J. Super. 270, 246 A.3d 279 (App. Div. 2021) · New Jersey Superior Court, Appellate Division
Holding
A criminal defendant was entitled to discovery of proprietary probabilistic-genotyping source code needed to challenge the reliability of the State's expert evidence, subject to protection for trade secrets.
Why it matters
A leading decision favoring meaningful adversarial testing when opaque software produces evidence used against a criminal defendant.
The court upheld admission of a TrueAllele probabilistic-genotyping likelihood ratio after a Frye hearing, while separately addressing confrontation and source-code arguments.
Why it matters
A major state high-court treatment of validation, general acceptance, and confrontation concerns surrounding algorithmically generated forensic evidence.