Every ruling that moved AI law.

Motions to dismiss, summary-judgment opinions, discovery orders, settlement approvals. Court, judge, date, and the holding — verbatim where possible.

Bartz v. Anthropic — Bifurcated fair-use order
NDCA·Alsup, J.·Jun 2025
Decided

Training on legitimately acquired books may be fair use; downloading and retaining pirated copies from shadow libraries is not. The order kept willful-infringement statutory damages alive and pushed Anthropic to a $1.5B settlement two months later.

HoldingPirated training corpus = not fair use
Bartz v. Anthropic — Preliminary settlement approval
NDCA·Alsup, J.·Sep 2025
Settled

Court preliminarily approved the $1.5B class-wide settlement: ~$3,000 per work to roughly 500,000 authors, destruction of pirated datasets, and a 120-day opt-out window. Final fairness hearing set for April 2026.

HoldingLargest U.S. copyright settlement
NYT v. OpenAI — Motion to dismiss
SDNY·Stein, J.·Apr 2024
Decided

Court denied OpenAI's motion to dismiss the core direct-infringement and DMCA §1202 claims. Time-barred theory rejected as to ongoing distribution. The case proceeded to extensive discovery and remains the flagship publisher case.

HoldingDirect infringement claims survive
NYT v. OpenAI — Log preservation order
SDNY·Wang, M.J.·May 2025
Decided

Magistrate ordered OpenAI to preserve all ChatGPT and API output logs covering ~20 million users — extending well beyond the standard 30-day retention. District Judge Stein affirmed in 2025. Set the privacy/preservation template for every output-infringement case that followed.

HoldingDiscovery preservation extended
Authors Guild v. OpenAI — Output infringement
SDNY·Oct 2025
Decided

Court held that ChatGPT's short, plot-level summaries of plaintiffs' novels may themselves constitute infringement, refusing to dismiss the output-based theory. The first decision in U.S. AI litigation to treat model outputs as actionable copies of copyrighted works.

HoldingOutput summaries may infringe
Getty Images v. Stability AI — UK High Court trial
UK High Court·Smith, J.·Nov 2025
Decided

After a six-week trial, the UK High Court rejected Getty's "model weights as infringing copies" theory and the secondary copyright claim, but found Stability liable for trademark infringement based on Getty watermarks reproduced in Stable Diffusion outputs.

HoldingWatermark trademark claims survive
GEMA v. OpenAI — Output infringement (DE)
Munich Regional Court·Nov 2024
Decided

German court ruled that ChatGPT's reproduction of song lyrics from GEMA's repertoire constitutes infringement, declining OpenAI's text-and-data-mining defense as to memorized output. Among the first European decisions to hold an LLM operator liable for generative output.

HoldingTDM exception does not cover output
Thomson Reuters v. Ross Intelligence — Summary judgment
D. Del.·Bibas, J.·Feb 2025
Decided

First U.S. summary-judgment ruling in an AI training case: Westlaw headnotes used to train a competing legal-research tool was not fair use. Non-generative AI, but the fair-use analysis is now cited in every books and news case.

HoldingTraining on competing data not fair use
Kadrey v. Meta — Partial summary judgment
NDCA·Chhabria, J.·2025
Decided

Court granted Meta partial summary judgment on a record limited to legitimately acquired Llama training material, while expressly leaving plaintiffs' DMCA §1202 and pirated-corpus claims for trial. Read alongside Bartz, the doctrinal split is sharpening.

HoldingFair use turns on corpus provenance
Walters v. OpenAI — Defamation dismissal
Ga. Super. Ct.·2024
Dismissed

First U.S. AI defamation case to reach a substantive ruling. Court dismissed: ChatGPT's hallucinated complaint naming the plaintiff was not "published with actual malice," and a reasonable reader would understand outputs are not factual reportage. Heavily caveated.

HoldingNo actual malice; not "publication"
Concord Music v. Anthropic — §1202 narrowing
NDCA·2024
Active

Court trimmed plaintiffs' DMCA §1202 (copyright-management-information) theory at the pleading stage but left core direct-infringement and contributory claims intact. The remaining claims are now pacing toward summary judgment.

Holding§1202 narrowed; core claims survive
Garcia v. Character.AI — Section 230 ruling
M.D. Fla.·Conway, J.·May 2025
Decided

Court rejected Character.AI's Section 230 defense to the wrongful-death and product-liability claims, holding that LLM outputs are first-party speech of the platform rather than third-party user content. The first U.S. ruling to deny §230 immunity to a generative AI company.

Holding§230 does not shield model output