Is opt-out a relevant tool for protection against AI?

The European Directive of April 17, 2019 introduced a series of exceptions to copyright law to accommodate text and data mining (TDM). One of its key features is the opt-out mechanism, which gives rights holders the ability to oppose the use of their works in data mining operations.

Since the rise of generative AI platforms, creative industries, particularly publishing, have expressed growing concern. The major issue: ensuring that copyrighted works are not used to train AI models without proper authorization from rights holders.

💡 Want to learn more about opt-out and its use in publishing? Explore our full 2025 guide on AI and the publishing industry. 💡

‍

A new exception within EU copyright law

‍

The 2019 directive introduced a new exception to the general framework of intellectual property, allowing for automated text and data analysis without prior permission from rights holders, as long as the rights holder hasn’t opted out.

Originally designed to support scientific research, the directive also authorizes TDM for broader purposes by private or public entities again, unless rights holders explicitly express their objection.

‍

How does opt-out work?

‍

The opt-out right allows authors and publishers to object to the use of their works for data mining. According to the directive, this opposition can be expressed:

• Without needing to provide a reason
• Through any appropriate means
• In the case of online content, via machine-readable formats, such as metadata or terms of service on websites

This gives publishers and other rights holders a way to technically and legally block unauthorized use of their works in AI training datasets.

‍

The limits of opt-out

‍

While opt-out may seem like a practical safeguard, it’s effectiveness is still largely theoretical. Most AI platforms do not disclose the data sources used in model training, making it nearly impossible for rights holders to verify whether their works were accessed.

Without a binding transparency requirement, there is no guarantee that opt-out declarations are respected.

‍

The role of the AI act

‍

A more robust solution may lie in the European Union’s AI Act, adopted in June 2023 and expected to take effect in 2025. The regulation includes specific provisions for generative AI systems, including:

• The obligation for AI platforms to provide summaries of the copyrighted works used in training datasets

This is a major step toward greater transparency, and it’s widely supported by the creative sector. In September 2023, a coalition of 76 organizations from the cultural and creative industries signed a public appeal calling for:

“A requirement for full transparency via a detailed list of works used by generative AI systems, including their sources made available to rights holders.”

‍

A divided Europe on transparency

‍

Despite support from the creative industries, some EU member states, including France and Germany, have expressed resistance to strict transparency rules, viewing them as potential obstacles to AI innovation in Europe.

Nevertheless, the AI Act could offer a turning point by creating a legal framework that both protects rights holders and fosters accountable AI development.

‍

In summary

‍

For now, the opt-out mechanism remains the primary legal safeguard for rights holders in the face of uncontrolled AI data scraping. However, its impact is limited without platform transparency and enforceable obligations.

The success of the AI Act will be critical in determining whether opt-out can function as a truly effective tool for protecting copyrighted works in the era of generative AI.

‍

‍

👋 Crealo is the best copyright management solution for publishers. Automate royalty reporting, contracts, all in one place. 👉 Contact us to learn more.

‍