Legislation proposing a risk-based framework for overseeing artificial intelligence applications has garnered strong backing from the European Parliament’s internal market (IMCO) and civil liberties (LIBE) committees. The committees endorsed the compromise, reached in lengthy three-way “trilogue” discussions with EU Member States at the close of the previous year, by a decisive vote of 71 to 8, with 7 abstentions.
The EU AI Act, introduced by the Commission in April 2021, outlines directives for AI developers based on the robustness of their models and the intended application of AI technology. It delineates prohibited uses of AI, such as social scoring, and provides guidelines for managing specific high-risk applications that necessitate adherence to data quality evaluation, threat assessment, and accountability in sectors like education, healthcare, and employment. Additionally, it mandates transparency requirements for tools and general-purpose AIs like deepfakes and IoT chatbots.
The majority of AI applications are likely to fall under the category of “low risk,” exempting them from the regulatory framework. To facilitate the controlled development, training, and testing of high-risk applications in a real-world setting, the proposal calls for the establishment of national governmental sandboxes.
While the Commission’s initial proposal for an AI code was initially met with minimal attention three years ago, the recent surge in conceptual AI technologies has propelled the program into the spotlight, triggering significant divergence among EU legislators. Despite a faction of Member States, spearheaded by France, advocating for an innovative AI regulatory structure to bolster domestic champions, other Members have pushed for amendments to ensure the legislation encompasses impactful public interest AIs.
Following arduous trilogue negotiations in December, a compromise text was agreed upon, encompassing provisions for general-purpose AI, which has faced ongoing resistance from certain governments. However, a recent Member State ballot on the negotiated text indicates a high likelihood of the EU moving forward with the proposed AI regulatory framework in the near future.
Although some procedural steps remain before the law’s full adoption, including a comprehensive vote by MEPs and final approval from the Council, potential conflicts between EU co-legislators seem unlikely. The resounding support from the political committees underscores the commitment of MEPs to secure the requisite majority for the law’s enactment and subsequent implementation later this year.
The phased implementation of the EU AI Act is anticipated to introduce stricter legal obligations for engineers within the scope between 2024 and 2027. While rules governing general-purpose AI, including governance, will come into effect nine months post-implementation, a code of practice will also be enforced. High-risk systems will have an extended 36-month implementation timeline, with the law potentially becoming fully applicable 24 months after enactment.
Despite the positive momentum towards the parliament’s full vote, some reservations persist. The Pirate Party, for instance, has expressed dissent, with MEPs labeling the Act as “flawed,” leading to a few council members voting against it. MEP Marcel Kolaja from the Pirate Party and IMCO commission highlighted concerns over the Act’s impact, despite acknowledging the European Parliament’s intentions.
Furthermore, MEP Patrick Breyer from the Pirate Party and LIBE committee cautioned against the potential ramifications of the EU AI Act, raising apprehensions about pervasive real-time physical surveillance and the normalization of high-tech monitoring practices across Europe. This critical stance underscores the ongoing debate surrounding the legislation’s implications on privacy and civil liberties in the region.
