EU data rules slow LLM rollouts, study finds

Global tech companies are increasingly pointing to the European Union’s regulatory framework when delaying or withholding new AI services in the bloc. A Center for the Governance of AI report covering 375 large language models released over the past 8 years (June 2018 – May 2026) found that, compared with the US, at least 11% of releases from companies including Meta, Google, OpenAI and Anthropic were delayed or never launched in the EU, while 7% were delayed or not released in the UK.

Regulatory factors were identified as the primary cause in 56 of 68 delay and non-release cases. Data protection rules were cited as the main barrier, especially for non-text systems involving images, audio or real-time video. The EU saw a 71-day delay for the Claude 3 Opus web app, while Meta had the highest overall rate, with more than a quarter (26%) of releases delayed or withheld in the EU and 15% in the UK.

The report points to stricter EU enforcement and slower clarification of how data protection rules apply to LLM training and deployment. The Digital Omnibus is being considered as a way to make data rules more workable for AI development, while reviews of copyright rules and AI Act provisions could further affect access to advanced models if applied too rigidly.