AI Act enters final phase of EU legislative process

The European Parliament adopted its position on the AI rulebook with an overwhelming majority on Wednesday (14 June), paving the way for the interinstitutional negotiations set to finalise the world’s first comprehensive law on Artificial Intelligence.

The AI Act is a flagship initiative to regulate this disruptive technology based on its capacity to cause harm. It follows a risk-based approach, banning AI applications that pose an unacceptable risk and imposing a strict regime for high-risk use cases.

“Is this the right time to regulate AI? My answer is resolutely yes. It is the right time because of the profound impact that AI has,” Dragoș Tudorache, one of the European Parliament’s co-rapporteurs on the AI Act, told his peers ahead of the vote.

AI Act’s plenary vote cast with uncertainty as political deal crumbles

The agreement among the leading groups in the European Parliament on the AI regulation is dead, opening the door for amendments from both sides of the aisle.

Banned practices

Where to draw a line on the types of AI applications that should be forbidden was at the centre of the last-minute attempts to change the text adopted at the parliamentary committee level.

The main point of contention related to Remote Biometric Identification. Liberal and progressive lawmakers sought to ban the real-time use of this technology but allow it for ex-post investigations on serious crimes.

By contrast, the centre-right European People’s Party tried to introduce derogations to the real-time ban for exceptional circumstances such as terrorist attacks or missing people. This last-minute attempt enraged the other political groups but was eventually unsuccessful.

However, a parliamentary official told EURACTIV that the point of plenary amendments is not always to modify the text but to send a political message. In this case, showing that the EU Parliament is not united on this sensitive issue aims to weaken the MEPs’ negotiating positions vis-à-vis the Council, which is pushing for broader law enforcement carveouts.

All other amendments outside the text agreed upon at the committee level were rejected during the plenary vote.

AI Act: EU Parliament walking fine line on banned practices

Members of the European Parliament closed several critical parts of the AI regulation at a political meeting on Thursday (13 April), but the prohibited uses of AI could potentially divide the house.

Foundation models & generative AI

The EU lawmakers introduced a tiered approach for AI models that do not have a specific purpose, so-called General Purpose AI, with a stricter regime for foundation models, large language models on which other AI systems can be built.

The top layer relates to generative AI like ChatGPT, for which the European Parliament wants to introduce mandatory labelling for AI-generated content and force the disclosure of training data covered by copyright.

With ChatGPT, generative AI caught mass attention, and the European Commission has launched outreach initiatives attempting to anticipate the AI rules and foster international alignment at the G7 level.

On these initiatives, leading MEP Brando Benifei warned that they “could become a context where the businesses will act to influence the legislative work,” which is now the focus of the lobbying efforts to water down the regulation. “But if we cooperate correctly between the institutions, this will be prevented.”

Benifei also told reporters that the Parliament wants to put on the table anticipating the two-year application period of the legislation, if not for all types of AI, at least, for foundation models or generative AI, given the disruptive effects these models are already having.

EU leaders race over outreach initiatives to anticipate AI rules

Top EU officials have set out several competing initiatives to engage with industry players and international partners before Europe’s new AI rulebook starts to bite.

Main changes

The MEPs introduced several other significant changes to the text, starting with the definition of AI aligned with the one of the OECD.

The list of prohibited practices was extended to subliminal techniques, biometric categorisation, predictive policing, internet-scrapped facial recognition databases, and emotion recognition software is forbidden in law enforcement, border management, workplace and education.

An extra layer was added for AI applications to fall in the high-risk category, whilst the list of high-risk areas and use cases were made more precise and extended in law enforcement and migration control areas. Recommender systems of prominent social media were added as high-risk.

The obligations of high-risk AI providers concerning risk management, data governance, and technical documentation were made more prescriptive. New requirements were introduced to conduct fundamental rights impact assessments and monitor environmental impact.

An AI Office was established to support coordination on cross-border cases, albeit the critical responsibility of settling intra-authorities disputes was left to the Commission. Tudorache’s view is to upgrade the AI Office into an all-compassing digital agency eventually.

AI Act moves ahead in EU Parliament with key committee vote

The European Parliament’s leading parliamentary committees have green-lighted the AI Act in a vote on Thursday (11 May), paving the way for plenary adoption in mid-June.

The AI Act is a flagship legislation to regulate Artificial Intelligence based on its potential …

Upcoming negotiations

MEPs will now enter interinstitutional negotiations with the EU Council of Ministers, representing European governments, and the European Commission in the so-called trilogues. The first trilogue will occur Wednesday evening after the EU Parliament’s vote.

However, this first session will be little more than a photo opportunity since, according to an agenda seen by EURACTIV, the EU institutions will merely state their positions and delegate the work at the technical level.

The negotiations will intensify once Spain takes over the rotating presidency of the Council in July, as Madrid has made finishing up the AI law its top digital priority.

The main points of contention are set to be on high-risk categories, fundamental rights, and foundation models. In contrast, issues such as governance, innovation and AI definition will likely be solved at the technical level.

The first operational trilogue is set to occur before the Spanish parliamentary elections. The Spaniards aim to reach a deal by November with three trilogues, but two have also been planned as a backup.

EU countries adopt a common position on Artificial Intelligence rulebook

EU ministers green-lighted a general approach to the AI Act at the Telecom Council meeting on Tuesday (6 December). EURACTIV provides an overview of the main changes.

[Edited by Nathalie Weatherald]

Read more with Euractiv

How Cyprus became the EU launchpad of Israel's spyware companies

Cyprus has become the privileged entry point for Israeli cybersecurity companies, including those operating in the shady world of surveillance software.