Three recent federal court decisions split 2-1 on whether AI training constitutes copyright fair use. Judges Alsup (Bartz v. Anthropic) and Chhabria (Kadrey v. Meta) found fair use; Judge Bibas (Thomson Reuters v. ROSS Intelligence) rejected it. But these results reveal deeper complexity about AI's impact on copyright markets.

In most cases, using copyrighted works to train AI models will likely be illegal due to concerns about market dilution. As explained by Judge Chhabria in Kadrey v. Meta, copyright holders generally should be compensated for AI training use.

Market dilution: It is a specific form of market harm recognized in copyright law’s fair use analysis. AI-generated content—created using copyrighted works as training data—reduces the demand, value, or visibility of the original human-authored works, even if the AI does not directly copy or regurgitate specific passages from those works. In other words, AI models can flood the market with similar but not identical works, thereby weakening the economic footing and discoverability of the originals. AI’s outputs are made possible precisely because they were trained on, and extract value from, the copyrighted works themselves via statistical inference. This kind of competition becomes problematic when enabled by unlicensed ingestion of protected works during training.

Consider a practical example: an AI model trained on 10,000 romance novels can generate hundreds of new romance novels daily at near-zero marginal cost. Even if these AI-generated novels don't directly copy specific passages, they flood the market with substitutes that compete for reader attention, visibility, and sales, reducing both the economic returns and discoverability of the original human-authored works.

Copyright holders' compensation: This is a crucial point.  Without compensation, the system dangerously leans on eroding the very economic foundations that incentivize authorship in the first place. The erosion of income—past and prospective—fundamentally undermines the economic justification for creating new works in the first instance. Licensing and compensation restore the balance by ensuring that the value extracted from copyrighted works during AI training flows back to creators, maintaining their financial viability and thus their ability to continue producing the very content that AI systems depend upon for training data. No race to the bottom should be allowed for another reason: it creates a self-destructive technological dynamic in which AI cannibalizes its own training data. 

How are things going in the European Union?

The United States operates under a flexible "fair use" doctrine that allows courts to determine on a case-by-case basis whether a particular use of copyrighted works is fair. This creates uncertainty but also adaptability.

By contrast, the European Union follows a rigid system of specific statutory exceptions enumerated in directives. Rather than a catch-all "fair use" standard, the EU provides detailed, limited exceptions: AI training falls under two specific text and data mining (TDM) exceptions included in the 2019 Copyright in the Digital Single Market Directive. The system centers on opt-out mechanisms: copyright holders must actively reserve their rights using machine-readable means, placing the burden on creators to prevent their work from being used in AI training.

The EU approach provides greater legal certainty through specific statutory rules but raises serious concerns about effectiveness. The system's reliance on administrative compensation rather than deterrent damages mirrors the EU's weak private antitrust enforcement, where the absence of treble damages has rendered private competition litigation largely ineffective compared to the robust US model.

Conclusions

The two jurisdictions are going in two different directions: a liability-based approach with potential massive damages exposure in the US, and a statutory remuneration scheme, with legal experts and policymakers proposing mandatory licensing and compensation regimes to balance AI innovation with creator compensation in the EU. Nothing new, therefore, because this reflects the divergent legal philosophies of deterrence between the US and the EU: the EU's centralized administrative control vs. the US's decentralized private empowerment.

CILC connects clients with qualified attorneys to navigate the complexities of multiple regulatory regimes, ensuring strategic advantage in a rapidly changing digital world.

ValueLicensing sells pre-owned Microsoft software licenses primarily to business and enterprise customers, including large organizations and public sector entities. Buyers typically save 30–70%, compared to new licenses, in the secondary market for pre-owned, perpetual Microsoft software licenses—especially for Windows and Office. 

Microsoft offers discounts or incentives to enterprises to surrender or not resell perpetual licences when they move to Microsoft 365 (subscription-based cloud products), and includes contractual restrictions in enterprise agreements to prevent resale, either by conditioning support or upgrades on not transferring old licences. Microsoft’s actions, if successful, would limit or eliminate the secondary market, forcing customers toward its own direct sales and cloud-based subscriptions.

ValueLicensing claims that Microsoft, by attempting to suppress the secondary market in pre-owned licenses, violated antitrust rules in the EU and the UK. Microsoft based its defense on arguments related to copyright law.

The Legal Framework

In both EU and UK copyright law, the principle of exhaustion (also known as the first sale doctrine) says that once a copyright owner has lawfully sold a copy of a work in the EEA/UK, its exclusive right to control the distribution of that copy is “exhausted." That means the purchaser can resell or transfer that copy without further authorization from the copyright holder. For software, this principle is codified in Article 4(2) of the EU Software Directive (Directive 2009/24/EC), which was retained in UK law after Brexit. The leading EU case, UsedSoft GmbH v Oracle International Corp (C-128/11, 2012), by allowing customers who buy a perpetual license downloaded from the internet to resell the license and accompanying copy, created the market for “used software” or “pre-owned licence.”

ValueLicensing’s business model relies squarely on that case and doctrine: it buys perpetual licences from corporate customers (e.g., firms migrating to the cloud) and resells them to others at lower prices. However, the UsedSoft doctrine applies only to computer programs, meaning that other categories of works, such as e-books, music, video, or art, are governed by different EU directives and for which the resale/exhaustion principle may not apply.

Microsoft's defense relies on this atomization of copyright law in the EU. Microsoft argued that not every element of Windows or Office is a “computer program” within the meaning of the Software Directive, so exhaustion doesn’t necessarily apply to the entire bundle. In particular, parts of the suite — e.g., graphical interfaces, fonts, help systems, or templates — may be literary or artistic works, or ancillary materials, not “computer programs.” Therefore, even if the executable code is subject to exhaustion, those other elements are not. If accepted, this argument would make it nearly impossible to run or resell a fully functional “pre-owned” copy without infringing copyright — effectively nullifying the UsedSoft precedent in practice.

Why is the case so relevant?

It directly affects the scope of the right exhaustion doctrine. If Microsoft’s view prevails, large vendors could effectively eliminate the secondary market for perpetual licences. If ValueLicensing wins, the decision would reinforce UsedSoft’s principle and potentially expand the lawful resale of legacy software — a major precedent for digital property rights and competition in software licensing.

CILC connects clients with qualified attorneys to navigate the complexities of multiple regulatory regimes, ensuring strategic advantage in a rapidly changing digital world.

The European Commission's Scrutiny under the DSA, the DMA, and the Antitrust Rules

Google’s new AI Overviews search summaries are under review by the European Commission (EC) for compliance with the Digital Services Act (DSA), the European Union Copyright Law, the Digital Markets Act (DMA), and the antitrust rules.

The multi-front EU regulatory review concerns the DSA’s safety and transparency rules, the DMA’s competition obligations, and traditional antitrust law. Google argues that AI Overviews increase website diversity and engagement, and that the safeguards in place mitigate content risks. 

The Digital Service Act

The EC is examining whether the new feature complies with EU rules for “Very Large Online Search Engines” (VLOSEs) under the Digital Services Act, which applies to platforms with over 45 million monthly users in the EU.

The concern: AI-generated summaries in search results might pose systemic risks to online safety, copyright protection, media pluralism, electoral integrity, and user rights.

Possible issues are AI “hallucinations,” spread of deepfakes, automated manipulation, and misinformation that could mislead voters.

For AI Overviews, additional concerns include the implications for copyright law. Many publishers and news organizations argue that Google’s AI Overviews are built using content scraped from their websites, which is then summarized and shown to users without acquiring licenses or obtaining authorization. However, LLM developers, including Google, may lawfully crawl and ingest copyright-protected works for model training if the works are made accessible and have not been reserved (opted out) by the right-holders. Current litigation will help define future boundaries.

The Digital Market Act

As a designated gatekeeper, Google must prevent self-preferencing, ensure fair access to search data, and avoid practices that could exclude competitors or disadvantage business users such as publishers. Complaints have arisen that AI Overviews have reduced referral traffic to news websites, impacting ad revenue and threatening media pluralism.

The concern is that the DMA requires business users to be able to opt out of features like AI Overviews without being delisted from Google Search altogether. European publishers and privacy groups have filed complaints, alleging that Google does not adequately enable this choice and that the integration of AI summaries unfairly shifts user attention away from original news sources.

The Antitrust Rules

The scrutiny under traditional antitrust rules primarily concerns whether Google has used its dominant position in general search to unfairly disadvantage publishers and competitors through its AI Overviews feature.

Both the DMA and traditional antitrust regimes apply to Google’s AI Overviews, but they differ fundamentally in their approach, process, and practical impact. The DMA sets out upfront, clear obligations for designated gatekeepers like Google, focusing on preventing specific unfair practices before they cause harm, which is irrelevant for the application of the rules. Antitrust operates through after-the-fact investigations based on complaints or regulators' concerns about conduct that has already occurred and caused harm.

What to expect

Burden on platform to show compliance is a central DMA/DSA feature; antitrust maintains a traditional regulator-plaintiff burden of proof. DSA and DMA both permit rapid, rules-based intervention with heavy fines and interim or structural remedies—while the antitrust process is reactive, lengthy, and focused on proven harm.

The regulatory convergence effort serves the broader goal of preserving media pluralism, consumer choice, and a level competitive digital market, as highlighted in recent EC and parliamentary statements.

Prepare for overlapping EU obligations and consider compliance strategies that cover multiple regulatory regimes.

At CILC, we provide expert legal guidance on antitrust, regulatory compliance, and intellectual property across the U.S., EU, and beyond. Our platform connects clients with qualified attorneys to navigate the complexities of multiple regulatory regimes, ensuring strategic advantage in a rapidly changing digital world.