Europe's legal framework for AI copyright is fundamentally different from the United States, and it poses a far more serious threat to technology companies developing large language models. Unlike the US, which allows "fair use" as a flexible defense for AI training, Europe has no such protection. This structural difference means that even if Meta, Google, or OpenAI win their copyright battles in American courts, they could still face devastating liability in Europe.

The stakes are enormous. As of June 2026, more than 70 AI copyright infringement lawsuits have been active or recently resolved in US and international courts. In May 2026, five major publishers, Elsevier, Cengage, Hachette, Macmillan, and McGraw Hill, filed a proposed class-action lawsuit against Meta Platforms, alleging that the company used millions of copyrighted works to train its Llama large language models. Two months later, three of those same publishers, along with author Scott Turow, filed a similar case against Google, claiming it obtained copyrighted books and journal articles, some from pirate sites, to train its Gemini models.

Anthropic became the first major AI company to settle these disputes, agreeing in 2025 to pay a group of authors $1.5 billion to resolve a class action lawsuit. OpenAI, Microsoft, and others continue to face similar claims from authors, news organizations, and visual artists.

Why Fair Use Doesn't Exist in Europe?

The fundamental legal difference between the US and Europe centers on how copyright exceptions work. In the United States, "fair use" is a flexible doctrine that courts apply by weighing four factors: the purpose and character of the use, the nature of the copyrighted work, the amount used, and the effect on the market for the original work. This flexibility has allowed some US courts to find that training AI on copyrighted material can qualify as fair use, though the case law is still developing.

Europe operates under an entirely different system. Under Dutch law and the EU framework established by the DSM Directive (Directive 2019/790), copyright limitations are exhaustively listed and narrowly interpreted. If a use doesn't fit within one of those specific exceptions, it's an infringement. There is no judicial discretion to recognize new exceptions based on technological necessity or public benefit.

The two text and data mining (TDM) exceptions in the DSM Directive are the most directly relevant provisions for AI developers. Article 3 covers research organizations conducting scientific research. Article 4 covers any user for any purpose, but rights holders can opt out by expressing a reservation in machine-readable form, typically through a robots.txt file. Once a valid reservation is in place, the exception no longer applies, and any subsequent scraping is infringement.

What Are the Key Legal Risks for AI Companies Operating in Europe?

German courts have already begun interpreting these exceptions in ways that threaten AI developers. In proceedings brought by GEMA (a German copyright collective) against OpenAI, the Regional Court of Munich held that where a model memorizes protected works and can reproduce them in outputs, the use goes beyond what the TDM exception permits. The case is now before the Munich Court of Appeal, and because Article 4 applies across all EU Member States, an appellate ruling against OpenAI would carry direct relevance for proceedings in other countries.

At the European Court of Justice (CJEU) level, a case called Like Company v. Google will address text and data mining in the context of generative AI directly. An Advocate General opinion is expected in September 2026, with a final ruling highly anticipated in 2027. This case could set precedents on highly disputed legal issues, though CJEU rulings rarely provide the comprehensive clarity that legal practitioners hope for.

Beyond copyright law, the EU AI Act adds a separate and reinforcing layer of obligations. The General-Purpose AI model obligations under the EU AI Act have applied since August 2, 2025, with enforcement fines applying from August 2, 2026, at up to 3% of global revenue.

Steps Technology Companies Should Take to Assess Their European Exposure

• Audit Training Data Sources: Companies should conduct a comprehensive review of where their training data came from, including whether any content was obtained from pirate sites or without proper licensing. The provenance of training data can give rise to independent liability, as established in the Bartz v. Anthropic case.
• Monitor Opt-Out Reservations: Rights holders are increasingly aware of the opt-out mechanism under Article 4 of the DSM Directive. Companies should regularly check for robots.txt files and other machine-readable reservations on websites they scrape, as violations of these reservations constitute clear infringement.
• Prepare for Collective Claims: The WAMCA (Web and Mobile Copyright Act) combined with developing case law on EU-wide jurisdiction suggests that large-scale collective AI copyright and privacy claims in the Netherlands may be a question of timing rather than probability. Companies should assess their exposure across European markets and consider legal strategies accordingly.
• Track Pending Case Law: The Thomson Reuters v. ROSS Intelligence case, which heard oral arguments on June 11, 2026, is the first US appellate consideration of whether training an AI model on copyrighted works constitutes fair use. Although a ruling in this case will not affect EU law analysis, European courts will read it with interest, and companies should monitor its outcome.

How Is the EU AI Act Reinforcing Copyright Protections?

The EU AI Act creates additional compliance burdens that work alongside copyright law to constrain AI development in Europe. General-purpose AI models, which include large language models like those developed by Meta, Google, and OpenAI, must comply with specific obligations under the Act. These requirements are separate from copyright liability but create a compounding layer of regulatory risk.

The combination of strict copyright exceptions, narrow TDM provisions, and the EU AI Act's general-purpose AI obligations creates a legal environment substantially more protective of rights holders than the US framework. A favorable ruling for AI companies in US copyright litigation would not protect them in European proceedings, making the European legal landscape the more consequential battleground for the industry.

Technology companies developing or deploying AI systems in Europe face a structurally different and considerably more protective legal framework than their counterparts in the United States. As pending cases like Like Company v. Google move through European courts, and as the EU AI Act enforcement mechanisms activate, the financial and legal stakes for the industry will only increase.