Europe’s AI Act Has Landed With Fewer Teeth Than Originally Proposed

European Union lawmakers on Wednesday approved the world’s most sweeping regulation so far of artificial intelligence. The AI Act, which will take effect in stages over the next several years, will ban some applications of AI that the law considers high risk, such as social scoring. It will also require the developers of AI models to adopt cybersecurity protections, among other safety and transparency requirements.

Europe has a reputation of being aggressive on the regulatory front, sometimes to the detriment of its tech and startup scene. This time around, though, the EU loosened up the AI rules after pushback from countries like France, Germany and Italy, who claimed that proposed restrictions would slow development by European large language model providers—such as France’s Mistral AI and Germany’s Aleph Alpha—without making AI safer. Lawmakers gave in and added exemptions from some of the act’s requirements for open-source models. 

That may have come back to bite regulators, though, as just last month, Mistral and Microsoft announced a partnership where Microsoft will make Mistral’s models available through its cloud platform, Azure. As part of that deal, Microsoft invested $16.3 million and took a minority stake in Mistral. And just this morning, data and AI company Databricks also invested in Mistral and will offer its models through its platform.