5 Things You Must Know Now About the Coming EU AI Regulation

Part One

If you’re a data analytics leader, you know artificial intelligence (AI) is key to getting the insights your business needs for digital transformation. You may even be aware of the progress made on the European Union’s draft Artificial Intelligence Act (EU AIA). Like the GDPR, this act will have wide-ranging implications for data-driven businesses — and like the run-up to the GDPR, it might seem as if there’s lots of time yet for planning and readiness.

And you could be very, very wrong. The time to start planning is NOW. Here’s what you need to know.

For details on each area of focus, see the Act draft text here. Image by the author.

1. How does the Act define AI? The definition is broad, encompassing machine learning approaches, logic- and knowledge-based approaches, expert systems, statistical approaches, Bayesian estimation, other statistical methods, and search and optimization methods. Advanced or augmented analytics and data science modeling likely will fall under the regulatory domain of this Act.
2. Are there financial penalties for violating the Act? Yes, and at this draft stage, they look substantial. Non-compliance can incur fines up to €30M, or 6% of worldwide annual turnover (whichever is higher). And providing incorrect, incomplete, or misleading information to the regulatory bodies can incur fines of up to €10M or 2% of turnover.
3. If my company is not doing business in the EU, will the Act apply? Very possibly. Like GDPR, the use of the output of an AI system in the EU can trigger penalties that apply extraterritorially to providers and users outside of the EU.
4. Which AI systems will likely be most scrutinized? AI systems that the Act considers high-risk seem to present the most significant exposure. Such approaches include those where AI is part of a safety component, is covered by EU single market harmonization legislation, is part of critical infrastructure, helps determine access to education or training, or impacts employment or access to financial services. Additionally, there are regulatory implications for government use of AI in law enforcement, border control, and the administration of justice.
5. What’s the anticipated timeline for the Artificial Intelligence Act becoming law? The Act has been assigned to EU Parliament committees, and the Council of Ministers is due to review the Act. Once the Act has passed, there will be a two-year implementation period.

While a two-year-plus timeline seems like ample time to prepare, the broad deployment of AI across modern enterprises, and the potential cost of fines, means ANY use of AI should drive preparation now.

See Part Two of this series for recommended early steps to prepare for the EU AIA, and subscribe for future updates.

Some resources:

CEPS Webinar: “A European approach to the regulation of artificial intelligence,” April 2021

Gartner Research: “How Forthcoming EU Legal Framework Will Affect Your AI Initiatives,” April 2020

The TIBCO Blog: A selection of posts on Artificial Intelligence

The Apex of Innovation (TIBCO): “Why Did Your AI Project Fail?”, September 2021