Gather — 27 May 2026 · Zeitgeist — a spike by Chris Gathercole

Elsevier vs Meta: First Science Publisher Sues Over Scraped Research Papers (Nature) — Elsevier joined the class action against Meta on May 11, 2026 over Llama training data. Science publishing entering the litigation: Elsevier has established licensing infrastructure and can demonstrate market harm from AI-generated scientific content that substitutes for licensed journal access — a materially stronger claim than individual author suits.

Part 3: Generative AI Training — US Copyright Office Report (Pre-Publication) (US Copyright Office) — Official position: AI developers using copyrighted works to train models that generate content competing with originals goes beyond fair use. The most authoritative policy statement on the training fair use question. Pre-publication — the final version will be the definitive document to track.

AI in Litigation: An Update on AI Copyright Cases in 2026 (Norton Rose Fulbright) — Tracks all major 2026 cases: OpenAI output logs ordered (January 5; 78M logs compelled March 9); Disney v. Midjourney; updated posture on all active suits. The output log discovery orders are the significant new development — courts are compelling AI companies to disclose specific outputs at scale, shifting legal exposure from training to output.
When Can AI-Generated Content Be Protected? Three German Rulings (Bird & Bird) — Three German court rulings in 2026 establishing thresholds for AI-generated content protection under German copyright law. First significant non-US jurisdiction case law on AI output copyright.

A New Look at Fair Use: Anthropic, Meta, and Copyright in AI Training (Reed Smith) — Covers both Bartz v. Anthropic (lawfully acquired = fair use; pirated = not) and Kadrey v. Meta in a single analysis. The $1.5B settlement and the sourcing-method distinction are the key facts; clearest single-source treatment of both cases together.

Open vs Closed AI Ecosystems Elsevier joining the Meta lawsuit (Llama specifically) confirms the IP exposure asymmetry: open-weight models face the same training-data liability as closed models but can’t negotiate licensing deals because weights are already distributed.
AI Impact on Society The US Copyright Office Part 3 position — AI-generated content competing with originals goes beyond fair use — will feed directly into the regulatory landscape as states and federal government develop AI legislation. Colorado AI Act (ai-societal-impact entry) includes provisions that intersect with this.
Claude Integrations Thomson Reuters v. ROSS (Third Circuit argument June 11) directly involves the same company as the Thomson Reuters CoCounsel MCP integration (claude-integrations entry this gather). The legal information sector’s simultaneous litigation and commercial partnership posture is a distinctive dynamic.

Emerging pattern: Output log discovery orders (78M OpenAI logs compelled, March 9) mark a doctrinal shift — courts are treating AI outputs as discoverable evidence, not just training data as the liability surface. The Morrison Foerster output-liability prediction (last gather) is materialising faster than expected. Training and output exposure are now both active.
Quality signal: The US Copyright Office Part 3 report is the most authoritative single document in the training-data fair use debate — an official government position that will influence courts, not just commentators. Monitor the final publication date; the pre-publication version may differ.
Keyword suggestion: "output discovery" AI copyright compelled 2026 — the output log discovery orders (78M compelled) are a new mechanism that will affect AI companies beyond OpenAI as other suits progress.