High quality data and high quality rights: consolidating Europe's AI ambitions with citizen rights

Are the EU’s ambitions of AI compatible with the rights of citizens? It all depends on how this is done and the political mindset to make it happen.

Artificial intelligence is becoming a cornerstone of the EU’s competitiveness strategy. For it to enable greater competitiveness, building world-class AI requires a regulatory environment that allows European developers to build and deploy models at scale.

Training competitive AI models that are truly “made in Europe” requires large, diverse datasets that reflect the continent’s diverse cultures, languages, and communities. Access to high-quality data is a prerequisite for this development; without it, models may fail to serve diverse populations or accurately diagnose medical conditions across different demographics.

However, datasets are typically drawn from the public internet which is frequently embedded with sensitive personal information. Special Category Data (SCD) – compiling information such as health conditions, religious beliefs, ethnic backgrounds, political opinions – carries the strongest legal protections under European frameworks like the GDPR. Yet it also seeps into the materials used for training large language models, placing SCD at the intersection of two founding commitments of the European project: the protection of privacy rights and the building of a competitive European economy.  

The European Commission’s Digital Omnibus takes a significant step in addressing this tension by introducing a new legal basis for the incidental processing of SCD during AI training, provided developers apply appropriate safeguards. But the design of that framework raises questions about what those safeguards should mean in practice – and whether the current proposals are up to the task.

The questions for policymakers are fundamental: what does meaningful protection of sensitive data look like in the context of AI training — and can it be achieved without sacrificing the diversity and quality of data that makes AI systems genuinely useful? What should the guardrails be for trustworthy by design and who is accountable for these? 

How the EU resolves these questions will matter both for the fundamental rights of European citizens and for the capacity of European developers to compete globally. This policymakers’ dinner will convene senior decision-makers, regulators, industry leaders, and civil society representatives for a frank conversation about what a workable and rights-respecting framework for SCD in AI training should look like.

Related content:

• Event | Privacy, plurality, prosperity: designing a workable Digital Omnibus
• Insight | Connectivity’s critical test: can Europe align ambition with delivery?
• Event | Simplify to amplify: unleashing innovation through a stronger Digital Single Market

Our events include photos, audio and video recording that we might use for promotional purposes. By registering you expressly confirm that you have read and understood Friends of Europe privacy policy. Should you have any questions, please contact us on privacy@friendsofeurope.org.

PHOTO CREDIT: Shutterstock | Krot_Studio