When it comes to AI, there is little agreement. Some see great potential, while others see great nightmares. Some see opportunities, and many see nothing but risks. In the EU, though, there is agreement on one thing, a new EU AI Law. In December 2023 the EU Parliament and Council agreed to a bill “…to ensure AI in Europe is safe, respects fundamental rights and democracy, while businesses can thrive and expand.”
Longtime compliance professional Letitia Adu-Ampoma (LinkedIn) explains that while the law won’t fully come into force for two years or more, it’s time for compliance teams to start paying attention and preparing.
The act is a part of the EU digital strategy, which is very focused on human-centric legislation. Its goal is to keep positive the impact of AI on people and society.
The approach it takes is risk-based, categorizing AI systems based on the level of risk: unacceptable (and prohibited), high risk, minimal risk and no risk. The act is very specific in how it defines which AI systems fall into each category. The unacceptable risk category, for example, includes social credit scoring, emotional recognition and behavioral manipulation.
Creators and users of high risk AI will be required to register the system in a public record. They will also need to conduct an impact assessment and be transparent.
Transparency will also be critical for generative AI. Providers will need to disclose the content generated and ensure that the models are not designed to create illegal content. There will also need to be governance in place to protect against copyright violations.
So what should compliance teams do now? Letitia recommends reading the guidance and to start preparing the business unit for what is to come.
Listening to the podcast would be good, too.
NOTE: This podcast was recorded in January 2024. The final version of the EU AI Act is yet to be released – a final EU parliament debate on the text will take place before its release. In the meantime, some ‘unofficial’ pre-final versions of the text have been leaked online in advance of this debate. The final EU definition of AI and key timescales for enforcement mentioned in the podcast are based on proposals made public. Listeners should look out for the final position which will be detailed in the EU AI Act when it is officially published in the next few weeks.