The Spanish Presidency is expected to complete technical preparations for the EU law on AI this week. Only then will it be possible to receive a revised negotiating mandate at what will probably be the last high-level meeting.
The AI Law is a legislative framework to regulate artificial intelligence based on its ability to cause harm. The bill is currently in the final stage of the legislative process, in which representatives from the EU Commission, Council and Parliament meet in so-called trilogue negotiations.
With Spain holding the rotating presidency of the EU Council, Madrid needs a revised mandate from EU states before the next trilogue on December 6. This could be the last opportunity to reach a political agreement on the AI law under the Spanish presidency – and before the disruption caused by the European elections.
The revised mandate will be divided into two negotiation packages, which the ambassadors will discuss on Wednesday and Friday next week. The Presidency is, therefore, trying to resolve all outstanding issues at a technical level by the end of this week.
Before a meeting of the Telecommunications Working Group, the Council’s technical body, the Presidency published an internal note, seen by Euractiv, on open source, base models and governance.
EU negotiations on AI law have stalled
A technical meeting on the EU law on artificial intelligence failed on Friday (10 November). Among other things, Germany called for the proposed approach to basic models to be withdrawn.
Powerful AI Models
Regarding the delicate issue of basic models, the document states that “the Presidency has done its best to coordinate the various comments in the Council so that an agreement with the European Parliament is possible”.
The published text is almost entirely identical to the Commission’s proposal, which Euractiv revealed on Monday. There is only one fundamental difference: the paragraph stating that the Commission can invite not only model suppliers but also national authorities and that civil society and other interested parties can support the process has been deleted.
The Spanish Presidency highlighted that the main element of giving a central role to codes of conduct, as requested by France, Germany and Italy, was included in the text – in the form of practical rules that would guarantee compliance with the regulation.
“Note that each base model rating scale will be updated in less than a year,” he continued.
An internal technical discussion on the topic is scheduled for Thursday (November 23). On Wednesday, the head of the Commission’s digital department, Roberto Viola, gave an hour and a half presentation to EU ambassadors on their commitment.
Euractiv understands from the document that France maintained its position of not regulating basic models during the meeting. Germany, on the other hand, was more flexible. A group of liberal countries expressed some reservations, but remained open to compromises in order to reach an agreement with the European Parliament.
At an internal meeting on Tuesday, several MEPs dealing with AI law criticized the Commission’s text regarding baseline models and governance. However, Euractiv believes that political pressure to finalize the law could persuade more skeptical MPs to accept the compromise.
A technical trialogue is scheduled for Friday. While no topics have been determined at the time of publication, Euractiv understands that foundational models, governance and law enforcement will likely be on the agenda.
Governance
The text on governance is the same that the European Commission shared with European Parliament co-rapporteurs on Sunday and that Euractiv reported on Tuesday. In the communication, the Presidency notes that the Council’s idea of a team of experts will be included in the final text, which implicitly refers to a scientific panel.
AI law: EU MEPs discuss AI control bodies
Members of the EU Parliament involved in drafting the AI law continue to discuss its implementation. Given recent proposals on the governance aspect of the law, there is much to discuss.
open code
With regard to open source, the Presidency argues that it is an essential driver of technological innovation. However, the risks associated with the use or characteristics of AI products do not depend on the type of license under which they are provided.
To strike a balance between innovation and protection, “the Presidency proposes an exemption from the provisions of the AI Act for systems and components provided under open source licenses.”
At the same time, the communication clarifies that high-risk AI systems and high-impact baseline models – or general-purpose AI models with systemic risks, as they are now called – would continue to be covered by the regulation. Member States were asked whether they agreed with this compromise.
Given that no legislative text has been proposed, it is not clear whether the Presidency will propose a new text or accept the European Parliament’s proposal. This also excluded AI components made available under free and open source licenses from the scope of the regulation. The exception is high-risk core applications and templates.
More unanswered questions
At a Telecommunications Working Group meeting on Tuesday, attachés discussed a number of less controversial aspects of the text. Euractiv understands that there were no red lines, although the Presidency’s room for maneuver was less clear in the law enforcement section, with only a small number of countries demonstrating flexibility.
National representatives were more flexible in user obligations, provisions on responsibilities along the value chain and application of the law in the new regulatory territory.
There was general agreement to align the definition with that of the OECD. There was more resistance to concessions to Parliament in assessing the impact on fundamental rights.
[Bearbeitet von Nathalie Weatherald/Kjeld Neubert]