Gather — 19 May 2026 · Zeitgeist — a spike by Chris Gathercole

The $1.5 Billion Reckoning: AI Copyright and the 2026 Regulatory Minefield (Complex Discovery) — Bartz v. Anthropic settled for $1.5 billion — the largest US copyright settlement on record. The fairness hearing was set for May 14, 2026. Judge Alsup found that Anthropic’s use of shadow library content (Books3, LibGen) was not fair use; the ruling forced the settlement rather than proceeding to trial on damages. Every AI developer is now repricing their training data risk accordingly.
AI IP Year in Review — First Federal Ruling Rejects Fair Use Defense for AI Training Data (Sterne Kessler) — Detailed analysis of Judge Alsup’s ruling: pirated/shadow library content is where courts are drawing the line. The same day, Kadrey v. Meta went the other way on lawfully-acquired data. The binary is crystallising: pirated training sources → not fair use; licensed/purchased sources → still contested but more defensible. Anthropic’s exposure was the specific sourcing method, not AI training per se.
Training Data or Taking Data? How AI Copyright Lawsuits Are Reshaping Creative Rights (BFV Law) — Full landscape survey: Bartz, Meta publisher suits, and the emerging framework courts use to distinguish lawfully-acquired from pirated training data. The sourcing provenance question is now the crux of the litigation — not whether AI training is transformative, but how the training data was obtained.

Fair Use and Artificial Intelligence 2026 Update (Ohio State University Copyright Resources) — Authoritative summary of the four-factor fair use analysis as applied to AI training: courts are consistently rejecting fair use for pirated content; for lawfully-acquired content the four-factor analysis still favours defendants in most circuits. The best single reference for the current state of the doctrine.
Court Rules AI Training on Copyrighted Works Is Not Fair Use — What It Means for Generative AI (Davis+Gilbert) — Analysis distinguishing the two categories: unlawfully-acquired (pirated) content → courts consistently refuse fair use. Lawfully-acquired content → courts remain more open. The headline “AI training is not fair use” is accurate but incomplete — the ruling is narrower than it sounds.
AI Copyright: Six Key Rulings (Norton Rose Fulbright) — The six most significant AI copyright decisions to date: training data fair use, output infringement, authorship, and the Supreme Court’s certiorari denial in Thaler v. Perlmutter. The most useful single-source case summary.

US Supreme Court Declines to Consider Whether AI Alone Can Create Copyrighted Works (Morgan Lewis) — Cert denied in Thaler v. Perlmutter (March 2, 2026): purely AI-generated works cannot be registered for copyright. The human-authorship requirement is now settled federal law. The output side of the equation has been decided; the training side is where the remaining uncertainty concentrates.

Court Rules AI News Summaries May Infringe Copyright (Copyright Lately) — Judge McMahon’s ruling: “substitutive summaries” — AI outputs that mirror the expressive structure and storytelling choices of source articles without literal copying — may plausibly infringe copyright. This expands AI liability from the training side to the output side in a way that affects RAG systems, summarisation tools, and any product that reads then rewrites copyrighted content.

Opt-Out Cop-Out? UK Government Rethinks Its Position on Copyright and AI (Lewis Silkin) — The UK abandoned its broad text-and-data-mining exception with creator opt-out following intense creative industry opposition. The mechanism was practically unworkable: creators couldn’t audit compliance, and the opt-out burden fell on individual rightsholders rather than AI developers.
UK Copyright and AI Report: The ‘Opt-Out’ Is Dead, But What Comes Next? (Reed Smith) — Four-strand work programme replacing the opt-out: consultation on digital replicas, a labelling taskforce with an autumn 2026 interim report, and a review of online rights management tools. The UK is now in a different policy lane from the EU’s transparency requirements and the US’s litigation-led approach.

California District Court Upholds Transparency Requirements for Generative AI Training Data (Norton Rose Fulbright) — AB 2013 (effective January 1, 2026) requires AI developers to publicly disclose training data sources, including synthetic data, copyrighted material, and personal information. The district court upheld it. California is now doing via transparency requirements what the UK tried via opt-out and the EU via risk classification.

AI Impact on Society The Bartz v. Anthropic settlement is the largest US copyright settlement on record — the financial scale is itself a societal impact story. AI companies are now pricing legal risk as a cost of doing business at the billion-dollar level.
Open vs Closed AI Ecosystems The pirated vs. lawfully-acquired training data distinction hits open-weight models harder — open-weight labs typically have less legal infrastructure for licensing at scale and more exposure to shadow library sourcing.

Quality signal: The Bartz v. Anthropic $1.5B settlement is the biggest single event in AI copyright history. Every AI training data strategy is being repriced against it. The pirated/licensed binary is now the operational distinction that matters.
Emerging pattern: Three different jurisdictions are now pursuing three different approaches: US litigation-led (Bartz/Meta lawsuits), UK transparency-plus-labelling, EU risk-tiered (AI Act GPAI provisions). A practitioner operating globally must navigate all three simultaneously.
Keyword suggestion: "substitutive summary" copyright AI output — Judge McMahon’s new framing covers RAG/summarisation output liability, a category that barely existed in case law six months ago.