Navigating the Frontier: New European Report Examines the Legal Challenges of Training AI Systems

A new report from the European Audiovisual Observatory highlights the critical legal tensions between artificial intelligence development and copyright protections. As the industry grapples with text and data mining exceptions and evolving transparency requirements, developers face mounting litigation that could redefine the “Fair Use” of creative content.

RMN News Legal Desk
New Delhi | July 28, 2026

The European Copyright Frontier in the Age of AI

The European Audiovisual Observatory has released a comprehensive IRIS report titled Copyright and AI training, authored by Senior Legal Analyst Diego de la Vega. This study arrives at a pivotal moment for Europe’s creative industries, examining the complex legal framework governing how copyrighted works are used to train large-scale AI models,.

Since the 2022 explosion of popular generative AI tools, the impact of these technologies across Europe’s business and creative sectors has been profound. However, the report identifies sensitive sectors where copyright has become a central concern, complicated further by the introduction of the landmark EU Artificial Intelligence Act and the Council of Europe Framework Convention on Artificial Intelligence.

  On Transparency: Enhanced clarity and a balance between dataset access and rightsholder protection are now the key priorities for European policymakers.

Text and Data Mining: The Heart of the Debate

At the center of the current legal uncertainty are Text and Data Mining (TDM) processes, which are essential for training AI. The report analyzes European legal provisions—specifically those rooted in the Copyright in the Digital Single Market Directive (CDSMD)—which allow for certain TDM exceptions.

However, significant doubts remain regarding whether these exceptions fully cover the breadth of modern AI training activities. The report highlights how approaches diverge across jurisdictions; for instance, the UK and other national frameworks vary significantly on matters of licensing, enforcement, and the rights of content creators versus developers.

Prompting and Platform Liability

Beyond the data used for training, the report explores the technical and legal status of “prompting”—the instructions users provide to AI systems. Using real-world examples like the GEMA v OpenAI ruling in Germany, the study breaks down the specific stages where copyright rights are engaged during model training.

Furthermore, the report examines how major platforms—including Adobe, ChatGPT, Claude, Copilot, and Midjourney—distribute copyright liability through their terms of service. These contractual provisions are increasingly vital as they define the balance of risk between AI providers and the end-users who generate content.

A Watershed Moment for Risk Management

The urgency of these legal questions is underscored by the recent $1.5 billion settlement between Anthropic and a class of authors. This landmark case signifies a turning point in AI governance, as the argument that web scraping constitutes “Fair Use” undergoes intense judicial scrutiny. For enterprise leaders and Chief AI Officers (CAIOs), evaluating a vendor’s exposure to model deletion and retrospective licensing costs has now become an essential part of risk management.

  On Judicial Scrutiny: With $1.5 billion settlements at stake, the era of unchecked web scraping for AI training is facing a severe legal reckoning.

As Europe awaits further guidance from its courts, the need for enhanced transparency and a clear balance between dataset access and rightsholder protection remains a top priority for future policymaking. A second report focusing on the copyrightability of AI-generated output is expected to follow in the second half of 2026.