The European Audiovisual Observatory released a report on July 27, 2026 that maps the legal friction between copyright and artificial intelligence training. Titled "Copyright and AI Training," the IRIS report arrives as generative AI tools pull from vast datasets of protected works, leaving creators across Europe uncertain about their rights.
AI now helps produce images, music, scripts and video with a few typed prompts. But the models behind those tools are trained on content often scraped from the web without explicit permission. The report examines how European copyright law is adapting to this new reality.
For creative professionals trying to understand these shifts, AI for Creatives provides training and resources focused on the intersection of AI and creative work.
Where the law meets the machine
The report's opening chapter charts the rapid rise of generative AI since 2022 and its effect on creative and economic sectors. It explains how generative and agentic AI systems create specific copyright problems when protected works are used for training. The analysis points to several sensitive areas and shows why copyright has become a central battleground for both creators and AI providers. Recent legal instruments like the Council of Europe's framework convention on AI and the EU's AI Act add further complexity.
The debate over text and data mining exceptions
The core of the legal debate is text and data mining (TDM). Chapter two analyzes the EU's TDM exceptions, first introduced in the Copyright in the Digital Single Market (CDSM) Directive. These exceptions allow content to be mined for training under certain conditions, but their relationship with national laws remains messy. Opt-out mechanisms, transparency obligations and the scope of the exceptions are all under scrutiny. The report shows how approaches differ between the EU, the UK and other jurisdictions, balancing rights holders' interests against developers' need for data.
As courts begin to rule on these issues-such as the GEMA v. OpenAI case referenced in the report-legal professionals are facing new terrain. AI for Legal offers insights for those handling AI-related copyright disputes.
Prompts, platforms and who bears the risk
Chapter three breaks down the technical steps of AI training where copyright can be implicated. It also clarifies the concept of "prompting"-the instructions users give to AI systems-and examines the legal standing of these prompts. A key question is whether a prompt provides enough human authorship to claim copyright in the output, a matter courts are just beginning to explore. Chapter four then looks at how platform terms of service divide copyright liability between providers and users. The report compares the terms of major platforms like Adobe Firefly, ChatGPT, Claude, Copilot and Midjourney, highlighting how transparency, opt-out options and contractual clauses directly affect both creators and end users.
Why this matters for creatives
The report's conclusions are clear: AI technology is advancing faster than the legal system, and training on copyrighted data remains a flashpoint. More transparency from AI platforms, clearer opt-out mechanisms and a better balance between data access and creator protection are the top priorities for future policy. For creative professionals, the rules governing how their work can be used for training are still being written. Monitoring platform terms of service, pushing for contractual protections and watching upcoming court decisions will be essential. A second part of the report, examining copyright protection for AI-generated outputs, is expected in the second half of 2026.
Your membership also unlocks:
