Negotiations over the AI Act enter the final stages in the European Parliament

Some progress was made this week on the AI Act as lead negotiators in the European Parliament came to an agreement on a set of compromised amendments. The EU AI Act, which was first published by the European Commission in April 2021, is a flagship proposal to regulate Artificial Intelligence technology in the EU. The overarching objective of the proposal is to ensure that AI systems placed on the EU market and used within the bloc are safe and respect existing laws on fundamental rights. 

This piece of regulation, which represents the world’s first attempt to regulate AI based on its potential to cause harm, is a key element of EU policy to foster the development and uptake across the Single Market of safe and lawful AI. 

The Commission’s proposal follows a risk-based approach and outlines a uniform, horizontal legal framework for AI that seeks to guarantee legal certainty. The regulation, which once adopted following inter-institutional negotiations, will be legally binding upon Member States (MS), seeks to promote investment and innovation in AI, enhance the governance and effective enforcement of existing law on fundamental rights and safety and will facilitate the development of a single market for AI applications. 

Negotiations over the file have been ongoing for some time now and while the Council of the EU adopted its so-called “General Approach” in early December, negotiations within the European Parliament under the helm of lead co-rapporteurs (negotiators), MEPs Brando Benifei (IT, S&D) and Dragos Tudorache (RE, RO) are ongoing. 

Two committees in the European Parliament are jointly responsible for the file – the Internal Market and Consumer Protection Committee (IMCO) and the Committee for Civil Liberties, Justice and Home Affairs (LIBE). With MEPs expected to finalise the text over the coming days, Mr Benifei and Tudorache are now trying to come to an agreement on the final pending issues. These include the list of AI uses that pose significant risks, the prohibited practices and definitions of the key concepts used in the draft regulation. 

The Council adopted its position on the Commission’s proposed text under the Czech Presidency on 6 December. In the Council text, MS came to a clear position on a number of controversial points. On prohibited AI practices for instance, the text extends to private actors the prohibition on using AI for social scoring and prohibits the use of AI systems that exploit the vulnerabilities of a specific group of persons, including those vulnerable due to their social or economic situation. In the European Parliament, the leading negotiators are proposing to expand the ban on social scoring not only of individuals but also to groups over inferred personal characteristics that could lead to preferential treatment. 

The definition of AI is a particularly sensitive topic in the proposal as it will determine the scope of the whole regulation. While it seems an agreement on this has yet to be reached by lead negotiators in the European Parliament, the Council’s text narrows the definition to systems developed through machine learning approaches and logic and knowledge-based approaches. This is to ensure the existence of sufficiently clear criteria for distinguishing AI from simpler software systems. 

All eyes will now be on IMCO and LIBE as the committees go to vote on their final text over in the coming weeks before it heads to Strasbourg for final approval from all of MEPs.