After extensive negotiations, the European Parliament, Commission, and Council came to a consensus on the EU Artificial Intelligence Act (the “AI Act”) on Dec. 8, 2023. This marks a significant milestone, as the AI Act is expected to be the most far-reaching regulation on AI globally. The AI Act is poised to significantly impact how companies develop, deploy, and manage AI systems. In this post, NM’s AI Task Force breaks down the key compliance timelines to offer a roadmap for U.S. companies navigating the AI Act. 

The AI Act will have a staged implementation process. While it will officially enter into force 20 days after publication in the EU's Official Journal (“Entry into Force”), most provisions won't be directly applicable for an additional 24 months. This provides a grace period for businesses to adapt their AI systems and practices to comply with the AI Act. To bridge this gap, the European Commission plans to launch an AI Pact. This voluntary initiative allows AI developers to commit to implementing key obligations outlined in the AI Act even before they become legally enforceable. 

With the impending enforcement of the AI Act comes the crucial question for U.S. companies that operate in the EU or whose AI systems interact with EU citizens: How can they ensure compliance with the new regulations? To start, U.S. companies should understand the key risk categories established by the AI Act and their associated compliance timelines.  

I. Understanding the Risk Categories
The AI Act categorizes AI systems based on their potential risk. The risk level determines the compliance obligations a company must meet.  Here's a simplified breakdown:

• Unacceptable Risk: These systems are banned entirely within the EU. This includes applications that threaten people's safety, livelihood, and fundamental rights. Examples may include social credit scoring, emotion recognition systems at work and in education, and untargeted scraping of facial images for facial recognition.
• High Risk: These systems pose a significant risk and require strict compliance measures. Examples may include AI used in critical infrastructure (e.g., transport, water, electricity), essential services (e.g., insurance, banking), and areas with high potential for bias (e.g., education, medical devices, vehicles, recruitment).
• Limited Risk: These systems require some level of transparency to ensure user awareness. Examples include chatbots and AI-powered marketing tools where users should be informed that they're interacting with a machine.
• Minimal Risk: These systems pose minimal or no identified risk and face no specific regulations. 

II. Key Compliance Timelines (as of March 2024):

In addition to the above, we can expect further rulemaking and guidance from the European Commission to come forth regarding aspects of the AI Act such as use cases, requirements, delegated powers, assessments, thresholds, and technical documentation.
 
Even before the AI Act’s Entry into Force, there are crucial steps U.S. companies operating in the EU can take to ensure a smooth transition. The priority is familiarization. Once the final version of the Act is published, carefully review it to understand the regulations and how they might apply to your AI systems. Next, classify your AI systems according to their risk level (high, medium, minimal, or unacceptable). This will help you determine the specific compliance obligations you'll need to meet. Finally, conduct a thorough gap analysis. Identify any areas where your current practices for developing, deploying, or managing AI systems might not comply with the Act. By taking these proactive steps before the official enactment, you'll gain valuable time to address potential issues and ensure your AI systems remain compliant in the EU market.

Anthony A. Laurentano also contributed to this article.