Three major regulatory shifts on AI training data and copyright (UK, EU, Italy)

News

The UK government withdrew a proposed broad copyright exception for AI training, instead committing to support industry-led licensing and global monitoring per its March 18, 2026 report. The European Parliament adopted a resolution on March 10, 2026 demanding stronger EU copyright protections for creators, transparency on AI training data sourcing, and application of EU copyright law to generative AI models regardless of training origin. Italy’s Court of Rome annulled a €15 million Garante fine against OpenAI on March 18, 2026, overturning the data protection authority’s November 2024 penalty for GDPR violations in ChatGPT training without full public reasoning.

Why it matters

These three developments reveal diverging regulatory trajectories with material consequence for AI training infrastructure and licensing frameworks. The UK’s withdrawal of a default opt-out exception signals a pivot toward voluntary, negotiated licensing rather than statutory compulsion—reducing friction for AI companies but requiring industry coordination. Conversely, the EU Parliament’s resolution pushes toward prescriptive territorial copyright enforcement and remuneration requirements, potentially constraining model training across EU borders and creating compliance complexity. Italy’s annulment of OpenAI’s fine on technical GDPR grounds (reasoning withheld) creates uncertainty around personal-data-driven training enforcement in Europe, even as the EU Parliament seeks stronger creator protections. Taken together, these actions show regulators scrambling to carve distinct policy positions on AI training rights before any international standard emerges—UK favoring market solutions, EU Parliament favoring statutory control—while GDPR enforcement remains tactically opaque.