AIIF Insights: Provisional EU AI Act
The December 9, 2023 passage of the EU AI Act was another AI law milestone with implications that extend to the global legal landscape.
It will be crucial for AI lawyers to understand the broader implications of the EU AI Act on international AI governance. Fortunately, Peter Craddock (Keller and Heckman) has selflessly offered to break it down for us from Brussels in an early March call. Look for that to be scheduled soon.
In the meantime, here are some key facts and implications of the provisional EU AI Act:
1. The First Comprehensive AI Law
The AI Act is being hailed as the first comprehensive global legal framework on AI. It will presumably anchor the global conversation toward its focus on safety, fundamental rights, and transparent, responsible AI development; with included measures like bans on social scoring, manipulative AI, and stringent rules for biometric identification systems. Its teeth are potentially sharp, with the proposed fines for non-compliance ranging from 7.5 to 35 million euros or 1.5% to 7% of global turnover, depending on company size and the violation.
2. High-Impact Generative AI & High-Risk AI
The AI Act categorizes certain AI systems as 'high-impact general purpose' and 'high-risk', requiring adherence to strict transparency standards and additional obligations such as risk management, incident monitoring, evaluation of AI models, and implementation of red teaming strategies.
3. However, Delayed Implementation and Provisional Nature …
Although significant, so very much yet remains to be seen from the enacting measures, the compromise Act is still merely a provisional agreement that may undergo further changes before the final wording is established and, even then, its enforcement is not imminent, with a timeline that extends two years post formal adoption by both the EU Parliament and Council.
4. … And Meanwhile
Two years in the AI landscape may be an eternity. Following on the U.S. Executive Order on AI, American executive rulemaking will have a wide opportunity to enter force first. Moreover, the Act’s provisional approval coincides with significant AI industry developments in Europe. For example, Paris-based AI startup Mistral released a new large language model with nothing but a torrent link, and a $415 million globally backed fund-raise.
5. EU as a Regulatory Force vs. EU as an Innovative Force
The Act underscores the EU's role as a key player in international technology regulation. A fine balance appears to have been achieved. Loud claims are being made that rules for large generative (or foundation) AI models are both: full of gaps and too strict to allow for European innovation.
For the sake of caution and given the time to its effectiveness, we might prefer that the AI Act start from a theoretically strict position. In that case, if AI innovation is weakened or AI is leaving the EU for other jurisdictions, good lawyering will be demanded to allow this domain to responsibly thrive in Europe.