Last week, a coalition of major music publishers including Universal Music Group, Concord Music Group and ABKCO filed a high-stakes federal copyright and DMCA lawsuit against AI company Anthropic. The publishers allege that Anthropic engaged in mass piracy of copyrighted musical works to develop and train its Claude generative AI models. The complaint, filed in the U.S. District Court for the Northern District of California and seeking damages reportedly exceeding US$3 billion, underscores the widening legal fault lines at the intersection of artificial intelligence and intellectual property rights.

While litigation over AI training data has become increasingly common, this case is notable for its emphasis not on abstract debates about “fair use”, but on something far more concrete: how the training material was obtained in the first place.

Allegations of Unlawful Acquisition

At the heart of the publishers’ claims are allegations that Anthropic acquired more than 20,000 copyrighted songs, lyrics and sheet music through unauthorised downloads from peer-to-peer “shadow libraries” using the BitTorrent protocol. The complaint argues that these works were subsequently incorporated into Anthropic’s training datasets, forming part of the foundation for Claude’s generative capabilities.

The publishers further assert that because BitTorrent operates through decentralised file-sharing, Anthropic did not merely download infringing copies but also distributed them to others. If proven, this would implicate not only reproduction rights but also the exclusive right of distribution, significantly increasing potential liability.

This is an important distinction for AI developers: copyright risk does not arise only from outputs generated by models, but also from upstream decisions about sourcing and handling data.

The Shadow of Bartz v. Anthropic

The lawsuit follows the earlier Bartz v. Anthropic case, where evidence emerged that Anthropic had torrented millions of files from pirate library websites. In that matter, a U.S. judge drew a critical legal line: while training on copyrighted material may, in some circumstances, be capable of falling within fair use, acquiring such content through piracy unequivocally is not.

The publishers’ current action builds directly on that reasoning. Rather than focusing solely on whether training itself is lawful, the complaint pivots toward affirmative allegations of unlawful acquisition and distribution, which are far harder to defend.

For rights-holders, this strategy is significant. It suggests that future claimants may increasingly focus less on theoretical arguments about AI training and more on tracing the provenance and legality of the underlying datasets.

DMCA Claims and Copyright Management Information

From a legal perspective, the inclusion of Digital Millennium Copyright Act (DMCA) claims adds further complexity. The publishers allege that Anthropic removed or altered copyright management information (CMI), such as attribution data or licensing metadata, in the process of ingesting works into training systems.

Under the DMCA, CMI violations can trigger substantial statutory damages independently of traditional infringement claims. For AI companies operating at scale, even technical handling of metadata may create exposure if rights information is stripped out during dataset processing.

This aspect of the claim serves as a warning: compliance is not only about whether content is licensed, but also about preserving the integrity of rights information throughout the AI development pipeline.

Why This Matters for UK and European Stakeholders

Although the case is being litigated in the United States, its implications are highly relevant for UK-based creators, publishers and technology businesses.

In the UK, the government’s attempted expansion of a broad AI training exception has met strong resistance from the creative industries, and current law remains comparatively rights-holder friendly. Text and data mining exceptions under the Copyright, Designs and Patents Act 1988 are limited, particularly where commercial use is involved and rights-holders have reserved their rights.

Meanwhile, the EU AI Act and evolving copyright policy continue to push transparency obligations around training data. Rights-holders across Europe are increasingly demanding disclosure of dataset sources, licensing arrangements, and opt-out mechanisms.

For London-based clients, this points to a growing convergence: even if AI companies train models abroad, the commercial deployment of those systems in the UK and EU markets will face increasing scrutiny.

Key Lessons for Creators, Rights-Holders and Innovators

This dispute offers several practical takeaways for businesses operating in creative and technology sectors:

1. Data provenance is now central to AI risk.
It is no longer sufficient to argue that machine learning requires broad access to information. Courts and regulators are increasingly focused on how that information was obtained.

2. Licensing strategy is becoming a competitive necessity.
Companies investing in AI should have documented workflows for acquiring and licensing copyrighted content, especially where creative works form the backbone of commercial products.

3. Creators should treat AI as both a threat and an enforcement opportunity.
Publishers and artists are becoming more sophisticated in pursuing claims not only for infringement, but for metadata stripping, unauthorised dataset building, and downstream commercial exploitation.

4. Governance and transparency are essential.
Robust internal compliance, clear audit trails, and contractual protections with data suppliers are quickly becoming standard expectations for AI companies seeking investment or partnerships.

Conclusion

The Anthropic lawsuit represents a pivotal moment in the evolution of AI copyright disputes. It reinforces a fundamental principle: innovation does not eliminate the need for lawful sourcing and respect for intellectual property rights.

For UK and EU rights-holders, it signals growing momentum toward accountability in AI training practices. For technology companies, it highlights the urgent need to align AI development with clear licensing, transparent governance and IP risk management.

Clients concerned about protecting creative assets, or about ensuring compliant use of third-party IP in AI development, should take note: a robust intellectual property strategy is no longer optional, but essential in the age of generative AI.