AI Copyright: Training-Data Lawsuits and Licences — 2026-10-03
A US federal appeals court has rejected fair use as a defense for AI training on copyrighted material for the first time, ruling that copying proprietary data to build a competing product constitutes infringement. The decision marks a major shift in AI copyright litigation and signals trouble ahead for music generators and other AI systems trained on protected works without consent. Meanwhile, the EU has launched consultations on whether new binding copyright measures are needed for generative AI.
AI Copyright: Training-Data Lawsuits and Licences — 2026-10-03
Top developments
US Appeals Court Rejects Fair Use Defense in Thomson Reuters Case
On September 29, the US Court of Appeals for the Third Circuit ruled against Ross Intelligence, an AI legal research tool, holding that training on Thomson Reuters' proprietary headnotes and using them to perform the same commercial function is copyright infringement—not fair use. The court backed by the Recording Industry Association of America (RIAA) and National Music Publishers' Association (NMPA) found that the AI company's copying of protected material did not qualify as transformative use. Critically, the court preserved fair-use arguments for generative AI systems, suggesting that general-purpose models trained on broad datasets may still have stronger defenses than narrow, competitor-targeting systems.
Bartz v. Anthropic Settlement Court-Approved; Fair Use Split on Training vs. Storage
The Bartz et al. v. Anthropic case, which received final approval on July 20, 2026, established a landmark $1.5 billion settlement and produced a nuanced court ruling: AI training on copyrighted books qualifies as fair use, but storing pirated copies in a centralized repository does not. This bifurcated ruling has become influential in ongoing litigation, including the New York Times v. OpenAI case, where both parties cite the distinction between training-stage copying and acquisition/storage questions. The ruling suggests courts are willing to permit transformative copying for model development while policing the unlicensed retention of infringing material.
Licensing Deals Accelerate; 36 Projected Deals for 2026, OpenAI Leads at 24
The market for AI content licensing is steepening: OpenAI has announced almost double the number of deals as Microsoft or Meta, with 24 publicly disclosed agreements and a pipeline projected to reach 36 total deals by end of 2026. Major publishers including News Corp (up to $250 million over five years covering WSJ, Barron's, MarketWatch, and international titles) and Apple (reported offers of at least $50 million to publishers including Condé Nast and NBC News) are licensing archives to AI companies. These deals signal that licensing is becoming standard infrastructure rather than an experimental sideline.
EU Launches Copyright Consultation on AI; Four Key Areas Under Review
On September 29, the European Commission opened public consultations testing whether new binding EU copyright measures are needed for generative AI, time-sensitive piracy, international phonogram remuneration, and scientific research. The consultation signals the EU's willingness to move beyond its existing Text and Data Mining (TDM) exception and craft AI-specific copyright rules. Stakeholders have flagged concerns that technical opacity should not be used to dilute enforcement rights creators already possess.
Copyright Alliance Tracks September 2026 Congressional Activity
The Copyright Alliance reported significant flurry of copyright-related activities in Congress ahead of mid-term elections recess, including two bills in House and Senate. While details remain preliminary, legislative momentum suggests policymakers are prioritizing AI copyright issues as courts render conflicting rulings on fair use and settlements reshape liability expectations.
Local view
Germany: A Tokyo court ruling (September 30) affirmed that voice mimicry via generative AI can trigger publicity rights protections, even when deletion was not granted. Meanwhile, German legal observers note that the GEMA v. OpenAI case (Munich Regional Court, November 2025) remains influential precedent: both storing and outputting copyrighted song lyrics without license violates German copyright law—a standard stricter than emerging US fair-use carve-outs.
Japan: A Tokyo District Court ruling (September 30) granted publicity rights recognition to a voice actor whose synthetic voice was used without consent, though deletion relief was denied. Japanese commentary emphasizes that Article 30-4 of the Copyright Act—the statutory exception for AI training—remains contested; recent suits by newspaper publishers signal that the opt-out versus opt-in framing of that provision will shape enforcement patterns going forward.
Context & numbers
- Thomson Reuters victory: First appellate-level rejection of fair use for AI training (September 29, 2026)
- Bartz settlement: $1.5 billion final approval July 20, 2026; training deemed fair use, but storage of pirated copies not
- OpenAI deal lead: 24 publicly announced licensing agreements vs. ~12 each for Microsoft and Meta
- Projected 2026 licensing deals: 36 total (vs. 0 in 2023, steep rebound from 2025 dip)
- Apple licensing offers: At least $50 million to major publishers for news archives
- News Corp deal ceiling: Up to $250 million over five years
- EU consultation scope: Four areas (generative AI, time-sensitive piracy, phonograms, scientific research)
On the radar
- NYT v. OpenAI docket watch: Federal judge in Manhattan must decide whether training large language models on copyrighted news articles qualifies as fair use; no ruling date set, but briefing complete
- Suno/music generator implications: Music industry stakeholders signaling that Thomson Reuters precedent may embolden suits against AI music tools; licensing may become mandatory
- UK copyright policy freeze: Government backed away from preferred broad text-and-data-mining opt-out approach (March 2026); further evidence-gathering underway before new legislation considered
- Copyright office guidance: US Copyright Office continuing to issue opinions on authorship and training-data liability; industry awaiting clarity on liability shield thresholds
Freshness check: All citations dated 2026-09-26 or later. Top story (Thomson Reuters, Sept 29) is 4 days old. Bartz settlement (final approval July 2026) included because it frames ongoing litigation cited in Sept 2026 briefs. Licensing data aggregated through July 2026 sources but referenced in Sept 2026 tracking reports.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
