The first phase of the European Union’s new AI Act recently went into effect. This law introduces new conditions under which companies are permitted to use AI. So make sure your company thoroughly familiarizes itself with the regulations to avoid hefty fines and other consequences. Companies must be ready for the first phase of the AI Act by February 1, 2025. In this blog, we’ll tell you how your company can prepare for that first phase!
How to implement the first phase of the AI Act
To help you get started with preparations for the first phase of the AI Act, we’ve outlined a brief step-by-step plan below:
- Step 1: Make an inventory of all the AI your business uses. In some cases, you might not even realize that you’re using a program that involves artificial intelligence. So be sure to look into this thoroughly. For example, AI might be part of the software you use, such as a CMS system.
- Step 2: What risk category does the artificial intelligence your company uses fall into ? To determine this, you can consult the European database, which lists all AI systems classified as high-risk. If you’re unsure, you can also ask the supplier of the software you use for assistance!
- Step 3: You now know which risk category each type of AI you use falls into, so you can immediately determine how you are and are not permitted to use artificial intelligence. Don’t just focus on the first phase; consider the subsequent phases as well. This way, you’ll be well prepared if you undergo an audit. The first phase of the AI Act applies only to the highest risk category.
- Step 4: Make sure you only use AI in ways permitted under the new AI Act to avoid fines and other penalties!
What types of AI fall into the "unacceptable risk" category?
To properly implement the first phase of the AI Act, it is, of course, important to know which types of AI fall into the category of unacceptable risk. Artificial intelligence falls into the highest risk group if it poses a direct threat to citizens. Consider, for example, things like social scoring, manipulating or misleading people—such as by using someone’s voice—biometric categorization, and predicting criminal behavior. If your company uses an AI application in any way that does any of these things, stop using that application. Or see if you can use the application in a way that is permitted.
What is the best way to phase out prohibited forms of AI?
If you discover that you are using certain types of AI that will be prohibited under the AI Act as of February 1, 2025, you will, of course, need to phase them out. This usually doesn’t happen overnight. There are a number of strategies you can use to do this:
- Stop using it: Stop using the AI application.
- Find a good alternative: There may be an alternative to the AI application that essentially does the same thing without using a prohibited form of AI. For example, if you have a newsletter program that uses social scoring, replace it with A/B testing!
- Replace prohibited AI with permitted AI: If your current AI uses some form of social scoring, you can choose to replace it with software that, for example, only considers certain characteristics of anonymous groups—such as age groups and geographic location.
- Replace the AI with human employees: Finally , you can choose to discontinue the AI application and have human employees take over the work. Perhaps you can also work within the same software without AI and have a person take over the AI’s tasks? It might take a little more time, but you won’t risk incurring hefty fines!
Conclusion
There are several ways you can properly prepare your business for the first phase of the AI Act, which takes effect on February 1, 2025. Whether you choose to discontinue your AI applications or look for alternatives, we recommend that you start the process early. It may still feel far off, but phasing out certain applications and work processes can take a lot of time. Before you know it, it’ll be February 2025 😉