Key facts
- A Governance AI study found that 11% of advanced LLM releases were delayed or blocked in Europe compared to the US.
- Regulatory factors, primarily data protection rules, are cited as the main reason for these delays.
- Meta had the highest rate of delays and non-releases in the EU, with over a quarter of its releases affected.
- The EU's enforcement of data protection laws, like GDPR, is seen as more aggressive than in the UK.
- The Digital Markets Act and the AI Act are also contributing factors to the delays in AI service rollouts.
A new study by Governance AI reveals that European Union data protection rules are significantly hindering the adoption and rollout of advanced large language models (LLMs). The report, which analyzed 375 LLM releases over eight years, found that 11% of these models experienced delays or were not released at all in the EU compared to the United States. Regulatory factors were identified as the primary cause in the majority of these instances.
Global tech companies frequently cite the EU's regulatory framework, including the Digital Markets Act (DMA), the Artificial Intelligence Act (AI Act), and existing data protection laws, as reasons for delayed service availability in Europe. The study highlights that data protection regulations are the most significant barrier, with non-text modalities like images and audio facing greater challenges than text-based services.
While the UK shares similar data protection laws with the EU, such as GDPR, the EU experiences larger barriers. This is attributed to the EU's more aggressive enforcement of these rules and a lack of clear guidance on how data protection applies to LLM training and deployment. The report notes that the full impact of the DMA and AI Act is yet to be seen as they are relatively new.
One of the report's authors, John Lidiard, a UK AI policy researcher at GovAI, emphasized the need for policymakers to balance regulatory implementation with the risk of delaying access to AI models for citizens and businesses. The EU is currently considering the Digital Omnibus to address data rule workability for AI development, but ongoing reviews of copyright provisions could further complicate future AI model availability.