On Wednesday December 6th, 2023, The European Parliament could not come to consensus on some of the remaining items to be discussed on the EU Artificial Intelligence Act (AI ACT), a horizontal legislation in development. Given the ambitious agenda, it is maybe not so surprising that it was not possible to come to an agreement. Items open for discussion were, amongst other things, the AI definition, a ban on exporting AI systems based on prohibited practices, biometric categorization and facial-scrapping databases.
One of the twist points at the latest moment concerns the rise of general-purpose AI services like OpenAI’s ChatGPT and Google’s Bard chatbot that had not been foreseen when the text of the AI Act was first drafted in 2019. The discussion on this subject had been ongoing, even till far after the official meeting had ended and after 36 hours of negotiations, finally an agreement was made late Friday evening.
The agreement does not include a complete ban on real-time face recognition systems as the European Parliament wished for, but they are only to be used by governments under strict conditions, where a judge needs to approve before such a system is used. The agreement does include ‘general purpose AI’ like ChatGPT, where manufacturers will need to be transparent on the data that was used to train the models. They also need to comply with copyright laws, and it must be exposed to the users when content has been generated using AI.
The AI Act is expected to be put to a vote at the European Parliament at the end of March. Because of the 2024 European Parliament Election on June 6 to 9, it was crucial that the agreement was reached, for after the election there might be shifts in parliament that could have potentially delayed the AI Act, and lots of negotiations would have needed to be redone. With only one meeting left in February, the AI Act was indeed on a crucial timeline, but given the current agreement it is likely that the deadline will be met. The AI ACT will apply two years after entry into force. See here for the official press release.
Although we have the agreement now, there are still a lot of uncertainties for European manufacturers on what will be required in order to bring their AI devices to the market, as many of the details are not yet known. Also unknown is the consequence regarding the biggest hurdle for medical device AI manufacturers; the acceptance of the use of standards already used in vertical legislation as the MDR and IVDR. The risk remains that, when the AI Act eventually will enter into force, the standards that are applicable under the AI Act will need to be used in addition to the standards on AI already used under the MDR and IVDR, which will lead to a huge additional burden and in addition will leave medical device manufacturers with conflicting requirements from both the AI Act and the MDR/IVDR.