Committees to vote on Compromise Amendments as AI Act moves through the European Parliament

Following months of tense technical and political meetings between relevant members of the European Parliament (MEPs) from the Committees for Internal Market and Consumer Protection (IMCO) and Civil Liberties, Justice and Home Affairs (LIBE), the European Parliament reached at the end of April a provisional political deal regarding its draft Report on the Artificial Intelligence Act (AIA). Members will now vote on the compromise text, which may still be subject to minor technical adjustments, at committee level in the following weeks and during the European Parliament’s plenary session in June. While last-minute high-level changes cannot be ruled out, the Parliament’s political groups are confident that the current text will pass after intense horse-trading on the most disputed issues. As the dossier moves towards forthcoming inter-institutional (“trilogue”) negotiations between the Parliament, Commission and Council, the most controversial points remain the provisions relating to biometric identification technologies and how to respond to the rise of generative AI, such as ChatGPT.

Biometric identification

Over the past weeks, the European Parliament has heavily debated the provisions on AI uses for biometric identification and categorisation of people. The latest Parliament’s compromise amendments propose the prohibition of the following AI systems:

• Real-time remote biometric identification systems in publicly available spaces (including by law enforcement);
• Post-remote biometric identification systems, unless subject to pre-judicial authorisation, or related to serious criminal offenses under Article 83 (1) TFEU;
• Systems using biometric traits to categorise people using, or inferring sensitive or protected attributes;
• AI systems that fill in facial recognition databases by indiscriminately scrapping face images from social media profile pictures, CCTVs.
• AI systems inferring the emotions of a person in the context of law enforcement or border management, or in the workplace and educational institutions.

In particular, the prohibition of post-remote biometric identification, heavily criticised by the conservative European People’s Party (EPP), has been ultimately included in the compromise text. This amendment reportedly passed in exchange for removing the proposal banning AI tools that monitor interpersonal communications.

General-purpose AI systems

The recent rise of generative AI systems, including ChatGPT, has sparked international debate around AI and become a touchstone issue for MEPs.

According to the Parliament’s compromise proposal, “general-purpose AI systems” are to be distinguished from “foundation models”, a sub-category of general-purpose AI systems subject to stricter requirements. Specifically, foundation models would be defined as AI systems trained on broad data at scale, designed for generality of output, that can be adapted to a wide range of distinct tasks. By contrast, general-purpose AI systems would be defined as AI systems that can be used in, or adapted to, a wide range of applications for which it was not intentionally and specifically designed.

Regarding foundation models, the Parliament envisions requirements such as testing and risk mitigation measures, data governance measures, increased transparency obligations, corrigibility, safety and cybersecurity requirements, amongst others. MEPs would include generative AI systems like, ChatGPT, in the category of foundation models. A last-minute amendment to the text requires providers of generative AI systems to design and develop the system in line with EU law and without prejudice to fundamental rights, including freedom of expression, and to document and make publicly available a summary of the use of training data protected under copyright law.

On the other hand, obligations for general-purpose AI systems would have to comply with high-risk application requirements according to Article 16 of the AI Act and with new value chain requirements. For instance, a downstream economic operator such as a distributor, importer or deployer of an AI system would be considered as the provider of a high-risk AI system and subject to relevant requirements in the following situations:

1. Put its name or trademark on a high-risk AI system already placed on the market or put into service;
2. Make a substantial modification to a high-risk AI system that has already been placed on the market or put into service;
3. Make a substantial modification to an AI system, including a general-purpose AI system which has not been classified as high-risk.

Next steps

While the Council already agreed on its internal position (“general approach”) in December 2022, the European Parliament’s s IMCO and LIBE Committees are expected to vote on their jointly drafted Report on 11 May 2023. A subsequent Parliament plenary session vote to adopt the final Report is currently expected for mid-June 2023. “Trilogue” negotiations between the Commission, the Council and the Parliament to agree upon a final text would then begin and can be expected to continue under the upcoming Spanish Presidency of the Council, which runs from June to December 2023. Once the final Regulation has entered into force, it would become applicable 24 months thereafter. This is currently expected to be in the first half of 2026.

For more information, contact Paolo Sasdelli

Sign up for our Connected newsletter for a monthly round-up from our Regulatory & Public Affairs team.