ChatGPT about Anthropic's filing: "Training fair use is strongest when paired with demonstrable, continuously improved technical and policy measures to prevent substitute outputs."

2025-11-28 · Source: Pascal’s Substack · Field: Legal & Regulatory — Intellectual Property & Patents, Compliance & Risk Management, Regulatory Affairs & Government Relations · Depth: Advanced, long

Summary

Anthropic's summary-judgment filing in the music publishers' lyrics case, dated April 21, 2026, attempts to bifurcate the legal dispute into "inputs" (training data) and "outputs" (user-generated content). Anthropic argues that training its Claude model on copyrighted lyrics, alongside billions of other works, constitutes "transformative" fair use because it serves a fundamentally different purpose than the original artistic expression. The company contends that market harm must be defined as substitution, not competition, and that licensing claims cannot be circular. For outputs, Anthropic asserts that direct infringement claims fail due to a lack of volitional conduct on its part, shifting liability to users under secondary liability standards. It also argues against DMCA claims, citing a lack of full work possession and intent. The most plausible near-term outcome is a partial win for Anthropic on training fair use, with continued litigation on outputs.

Key takeaway

For legal and product teams developing large language models, understanding the distinction between training data ingestion and model output generation is critical. Your defense against copyright infringement claims will likely hinge on demonstrating the transformative nature of your training process and establishing user volition for any infringing outputs. Prioritize implementing strong technical and policy guardrails to prevent verbatim reproduction and prepare for ongoing appellate scrutiny of fair use in AI.

Key insights

AI training on copyrighted works is argued as transformative fair use, distinct from potentially infringing model outputs.

Principles

• Training use is transformative if it serves a different purpose than the original work.
• Market harm requires substitution, not just competition from new works.
• Direct infringement requires volitional conduct by the alleged infringer.

Method

Anthropic's legal strategy separates AI training (inputs) from user-generated content (outputs), asserting fair use for inputs and shifting output liability to users, while challenging market harm definitions and DMCA claims.

In practice

• Implement robust anti-regurgitation measures in AI systems.
• Document non-infringing uses of AI models.
• Focus on user-instigated conduct for output liability.

Topics

• AI Copyright Litigation
• Fair Use Doctrine
• LLM Training Data
• Anthropic
• Copyright Infringement Liability

Best for: CTO, VP of Engineering/Data, Executive, Legal Professional, Policy Maker, Director of AI/ML

Related on AIssential

• Anthropic Code Crisis Creates Copyright Contradiction — Anthropic's DMCA takedown for leaked code contradicts its prior legal arguments on fair use for AI training data.
• Fair Use as an Industrial Policy: What 'AI Progress' Is Really Arguing For — and What It Leaves Out. Critics can point to any counterexample—model outputs that substitute for works... — Fair use is presented as a national policy engine for AI innovation, crucial for U.S. technological leadership.
• The Atlantic uncovers millions of copyrighted songs in AI training data — Millions of copyrighted songs are embedded in AI music training data, fueling ongoing legal disputes over fair use.
• Cognella v. Anthropic: Anthropic allegedly acquired pirated books from shadow libraries, torrented them, redistributed them through peer-to-peer networks, stripped copyright management information... — AI copyright litigation is shifting from abstract "training" debates to concrete allegations of piracy and illicit content acquisition.
• Microsoft trained its MAI models on unlicensed web data despite promising "enterprise grade, clean and commercially licensed data" — Microsoft's MAI models used unlicensed web data, challenging its "clean data" claims amidst ongoing fair use legal disputes.
• Microsoft built supercomputer to help OpenAI infringe copyrights, NYT alleged — New legal precedent requires plaintiffs to prove intentional inducement for contributory copyright infringement in AI training cases.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.