Share This Article
The European Commission’s recently unveiled disclosure template under the upcoming EU AI Act aims to bring clarity to how AI models are trained with implications for copyright holders and individuals’ privacy rights.
Below, we delve into what the new disclosure template means, why it’s stirring controversy, and how it could reshape the entire AI landscape in Europe and beyond. You can listen to the episode of the podcast on the topic below and on Apple Podcasts, Google Podcasts, Spotify, and Audible and read the article below:
The New Copyright Disclosure Template under the AI Act
The European Commission’s AI Office has introduced a draft template that requires general-purpose AI model providers to disclose a “sufficiently detailed summary” of their training data. While it may sound like a straightforward procedural shift, the ramifications could be transformative.
- Breadth of Coverage
- The template applies to multiple stages of an AI model’s lifecycle, from the initial pre-training phase to subsequent fine-tuning. This means that if a company first trains its model on a large, general dataset and later refines it with specialized industry data, both steps could fall under the purview of these new disclosure obligations.
- Data Volume per Modality
- Companies will be asked to itemize how much data they’ve used for text, images, audio, video, and other data “modalities.” This includes specifying the type of content—like music, film segments, or advertisements—within those datasets. For AI developers, that level of detail can be challenging to produce and may even risk revealing trade secrets.
- Source Listing
- One key requirement is the disclosure of data sources. Simply stating “collected from the internet” won’t suffice. AI developers must identify their largest data contributors, offering more transparency—and potentially more ammunition for right holders seeking to protect their works.
- Copyright Compliance
- Perhaps the most sensitive area is how AI companies address copyrighted content. The template demands an explanation of how developers comply with the EU Copyright Directive, particularly through the text-and-data mining (TDM) exception.
Why Copyright Disputes Loom Large
Copyright holders stand to gain a powerful new tool: a roadmap to see whether their works have been used without permission. Even though some uses of copyrighted material for the purpose of AI training may be shielded by the TDM exception under EU law, the scope of that exception is limited.
- Rights Holders’ Perspective: They argue that detailed disclosures will enable them to protect their reservation rights, ensuring that if their works are being leveraged by AI providers, they are fairly compensated. They also worry that unlicensed usage could eat into potential revenues and undermine creative industries.
- AI Developers’ Concerns: On the flip side, AI companies often rely on vast datasets compiled from a range of public and sometimes proprietary sources. Mandating granular details about those sources could hamper innovation, reveal competitive secrets, and introduce administrative burdens that slow down research and development.
- Not Just for Developers: Importantly, the impact doesn’t end with the companies that build AI systems. Businesses that integrate AI tools into their workflows also face potential liability if the underlying models were trained on improperly sourced data. For instance, a marketing agency using an AI-driven content creation tool could, in theory, be implicated if the model’s training set contained copyrighted materials outside the TDM exception.
Text-and-Data Mining Copyright Exception: A Partial Shield
The EU’s text-and-data mining exception allows for certain uses of copyrighted material without explicit consent, primarily for research or nonprofit purposes. It’s a critical legal pathway that AI developers often rely upon to gather and analyze massive amounts of data quickly.
However, this exception has well-defined boundaries:
- It doesn’t universally protect commercial applications.
- It generally requires that right holders have not reserved their works from being mined.
- It typically applies to specific research or innovation contexts, leaving many commercial endeavors out of scope.
Because of these limitations, AI companies and the businesses adopting their solutions may find themselves navigating a legal minefield—especially when faced with right holders who argue that their content was used in training data against the spirit of the directive.
Balancing Transparency and Trade Secrets
One of the most contentious aspects of the new template is how much detail is enough—and how much is too much. Many developers argue that disclosing top data sources in granular detail essentially hands over a roadmap of their competitive edge. The risk is that proprietary or strategic information could become public knowledge, reducing incentives to innovate.
On the other hand, regulators and advocates for transparency maintain that basic disclosure is vital to protect privacy and intellectual property rights. By understanding where training data comes from, affected parties—whether they are authors, composers, or data subjects—can better exercise their legal protections.
This tension is poised to be a central talking point in the ongoing consultation and lobbying process. If the EU requires “maximum” detail, it might deter smaller or cutting-edge AI developers who fear losing their competitive advantage. If the disclosure bar is set too low, however, right holders and the public may feel the regulation lacks teeth.
Looking Forward: Possible Outcomes
- Refined Guidelines
- The European Commission may issue more granular instructions on the level of disclosure required, potentially striking a compromise between transparency and proprietary secrecy.
- Voluntary vs. Mandatory
- The final form of this template could either remain a strong recommendation under a code of practice or become enshrined in mandatory regulations. Developers and businesses will need to be prepared for either scenario.
- Legal Precedents
- Early litigation could shape how strictly courts interpret these obligations. A successful lawsuit by right holders might embolden others, potentially triggering a wave of copyright-related disputes.
- Impact on Global AI Regulation
- The EU often sets the tone for technology regulations worldwide. If the template and accompanying rules become a benchmark, other jurisdictions may follow suit or develop similar frameworks tailored to their legal systems.
Conclusion
The European Commission’s new disclosure template for AI training data embodies a pivotal shift in how we think about transparency, intellectual property rights, and accountability in artificial intelligence. While it aims to empower individuals and right holders alike, it also places an unprecedented level of scrutiny on the data practices of AI developers and users.
Whether you’re creating AI applications or merely using them in your daily operations, these regulatory changes should be on your radar. They could redefine competitive strategies, risk profiles, and even the future shape of Europe’s AI industry.
Stay tuned for more updates on the EU AI Act and read more articles on the topic HERE.
