On 6 December, the Telecommunications Council of the EU convened to formally adopt its General Approach on the Artificial Intelligence (AI) Act. The formal approval of the text follows in the wake of a COREPER I meeting that took place on 18 November, during which the EU ambassadors provisionally greenlighted the Czech Council Presidency’s final compromise text. The AI Act, which was proposed back in April 2021, is the European Commission’s first ever attempt of establishing a horizontal regulation on AI. With the proposal, the Commission aims to lay down harmonized rules on AI through a risk-based approach, characterised by being human-centred, safe and trustworthy for both people and businesses.
With the Member States’ final rubber-stamp of their position on the file, the Council now has its mandate for the upcoming trilogue negotiations with the European Parliament and the Commission. Some of the most crucial elements of the Council’s General Approach include wording on how to define an AI system, classification of and requirements for high-risk AI systems, prohibited AI practices, general-purpose AI systems, regulatory sandboxes, etc.
Regarding the highly controversial topic on how to define AI systems, the Council’s text provides a narrower definition than the one proposed by the Commission, defining AI systems as “machine learning and logic- or knowledge-based systems.” As such, the Council seeks to ensure greater legal clarity by distinguishing AI from simpler software systems. On high-risk AI, which similarly has caused heated debate among EU institutions and stakeholders, the Council’s General Approach adds a horizontal layer on top of the high-risk classification. The rationale behind this is to ensure that AI systems that are not likely to cause serious violations to the fundamental rights of users, or other significant risks, are not captured by the regulation. Ecommerce Europe is positive towards this addition to the text, as we previously have raised concerns about the Commission’s proposal, especially regarding the many AI systems that inadvertently could be classified as high-risk, and thus be subject to burdensome high-risk requirements, simply because the AI system is used in connection with certain high-risk products or industries. Maintaining the original wording of the Commission’s proposal would be detrimental to the incentives for innovation in the field of AI, and we are therefore pleased to see that the Council seeks to tackle this issue in its text. Moreover, the Council text also clarifies many of the requirements for high-risk AI systems, making them less cumbersome for stakeholders to comply with, while also making them more technically feasible, for instance regarding the quality of data, or the technical documentation to be provided by SMEs to demonstrate that their high-risk AI systems fulfil the requirements. Furthermore, the Council’s General Approach takes into account the fact that AI systems are developed and distributed through comprehensive value chains, including several actors, for which reason the text provides more explicit wording on the responsibilities and roles ascribed to these different actors. Similarly, the text also clarifies the relationship between responsibilities under the AI Act and responsibilities laid down in existing Union legislation, such as data protection or sectorial legislation. In terms of general-purpose AI, the Council position remains that the Commission, by means of an implementing act, should formulate a list within 1.5 years after the entry into force of the regulation on the requirements that general-purpose AI systems should comply with.
Finally, the Council text also provides clarity on elements related to regulatory sandboxes e.g., by adding that development and testing of AI systems in a controlled environment could also take place (unsupervised) in real world conditions under specific conditions and safeguards. The text also adds provisions on innovation support activities to be facilitated in the Member States, such as the organisation of trainings to explain the application of the AI rulebook to SMEs, start-ups and local authorities with a view to provide more clarity on the legal landscape surrounding AI.
Despite the Council’s formal adoption of its General Approach, the European Parliament’s responsible Committees, IMCO and LIBE, are still in the process of finding common ground on the file, and the vote in the Committees is thus not expected until spring 2023. As such, Ecommerce Europe will continue to monitor the negotiations on the AI Act, which are likely to continue throughout 2023.
If you have any questions or wish to know more about the topic, please feel free to contact us at email@example.com.