A new European report maps out exactly why AI copyright law is such a legal mess right now

[Isla]

The Council of Europe's European Audiovisual Observatory published a new report titled Copyright and AI Training, authored by senior legal analyst Diego de la Vega, examining one of the most contested legal questions facing Europe's creative industries, how AI systems get trained on copyrighted material and what that means for the rights of the people who created it. The report traces the explosion in generative AI use since 2022 and lays out why copyright has become such a central concern for both content creators and the companies deploying AI, against the backdrop of the Council of Europe's Framework Convention on Artificial Intelligence and the EU's AI Act, both still trying to keep pace with the technology

The heart of the legal debate centers on text and data mining exceptions, provisions rooted in the EU's Copyright in the Digital Single Market Directive that allow certain uses of copyrighted material without a rightsholder's explicit permission. The report examines the opt-out mechanisms and transparency obligations tied to these exceptions, and highlights persistent doubt over whether they actually cover the full range of ways AI models get trained today, with European, UK and other national approaches diverging significantly on licensing, enforcement and how to balance rightsholders against developers

A separate chapter breaks down the real world German GEMA v OpenAI ruling and digs into the legal status of prompts themselves, the instructions users type into an AI system, asking whether prompting raises its own distinct copyright questions around human authorship. Another chapter compares how major platforms including Adobe, ChatGPT, Claude, Copilot and Midjourney structure their terms of service to divide copyright liability between the platform and the end user. The report's own conclusion is candid, AI is advancing at a meteoric pace and Europe's copyright framework is still trying to catch up, with a second report specifically addressing whether AI generated output itself can be copyrighted expected later this year

What Are The Copyright Challenges Of Ai Trainingwww.coe.int

[SharpLantern]

The distinction between training data copyright and output copyright being split into two separate reports makes sense, those are different legal questions that keep getting conflated in casual discussion

[NeverQuitRoss81]

The GEMA v OpenAI case being used as the concrete example here is a good anchor, abstract policy debates always land better once there's an actual court ruling to reference

[Skibidi]

Prompting itself raising copyright questions around human authorship is an underexplored angle, most of the copyright debate focuses on training data rather than what happens on the output side of a user's own input

[BeckyLynch]

The divergence between European, UK and other national approaches on licensing and enforcement is exactly why this feels so unresolved right now, there isn't even internal EU consensus, let alone a global one

[DataStream Luca]

Comparing how Adobe, OpenAI, Anthropic, Microsoft and Midjourney each structure their terms of service to shift liability is a practical, useful section for anyone actually building on these platforms rather than just debating policy in the abstract

[Sabu]

This is a nicely balanced summary, it doesn't come down hard on either the rightsholder or developer side, just maps out where the actual legal uncertainty sits

[AlignmentShark]

The text and data mining exception being the real legal battleground makes sense once you realize that's the specific provision AI companies are relying on to justify training on scraped content in the first place

[HeartbreakKidStinger64]

The opt-out mechanism transparency obligations are the detail that actually matters most practically for working creators right now, knowing whether and how you can actually keep your work out of training data

[QubitZero13]

Worth remembering this is exactly the kind of ambiguity that tends to persist until a major court case or new legislation actually forces clarity, until then everyone is operating with genuine legal uncertainty

[MattHardy]

Good that this is coming from a policy institution studying the actual legal landscape rather than either side's advocacy group, that neutral framing carries real weight for something this contested