EU lawmakers to discuss AI rulebook’s revised governance structure

1 year ago 43

The Members of the European Parliament involved in the AI law will discuss the governance aspect of the legislation in light of the recent discussions on the most powerful AI models on Tuesday (21 November).

The AI Act is a landmark law that regulates artificial intelligence based on its capacity to cause risk, currently at the last phase of the legislative process, where the EU Parliament and Council, the co-legislators, agree on the final provisions and the Commission acts as the ‘honest broker’ between the two.

The Commission drafted the compromise text shared with the European Parliament’s co-rapporteurs on Sunday, including establishing a new proposal on regulating general-purpose AI models, reported by Euractiv on Monday. The text includes some suggestions from the leading MEPs.

As noted by the co-rapporteurs in a note to the margin of the text, a crucial article in a previous iteration of the text allowed the European Commission to take measures in case of widespread infringements has disappeared.

AI Office

According to the new text, the Office oversees the enforcement of General Purpose AI (GPAI) models, while EU countries should facilitate its tasks.

In particular, the AI Office is to monitor and request any relevant information for applying the AI Act’s dispositions on GPAI model providers, including their adherence to approved codes of practice.

Downstream economic operators who built a system on top of a GPAI model are given the right to lodge a reasoned complaint for alleged infringements of the AI Act.

Additionally, the AI Office is empowered, after consulting the Board, to conduct ex-post evaluations on GPAI models to assess their compliance with the AI regulation when the gathered information is inconclusive or to investigate systemic risks at the EU level.

The Office might appoint independent experts from the scientific panel to evaluate on its behalf. In these cases, the Office might request access to the models via an Application Programming Interface (API) or other technical means.

For the leading lawmakers, the access request should be preceded by a structured dialogue between the AI Office and the model provider.

Moreover, the AI Office can request GPAI model providers to take appropriate measures to comply with their obligations or to implement mitigation measures where there are substantiated concerns of systemic risk at the EU level, including restring its availability or withdrawing it from the market.

If, during a structured dialogue, the GPAI model provider offers commitments to implement relevant mitigation measures, the AI Office might make these commitments mandatory and refrain from any further action.

A sanction regime for violations of the AI Act and failing to cooperate with the AI Office is mentioned but with no reference to the maximum percentage of the company’s global turnover. The modalities for adopting the fines are to be defined in secondary legislation.

The co-rapporteurs also want the AI Office to provide coordination support for joint investigations.

European Artificial Intelligence Board

For the compromise, the Board will comprise one representative per EU country appointed for three years, renewable once. The leading MEPs want the European Data Protection Supervisor, the Fundamental Rights Agency, and the EU Cybersecurity Agency ENISA to participate as observers.

The Board must adopt its internal rules by a two-thirds majority and have two standing subgroups with market surveillance and notifying authorities, respectively. The AI Office is to provide the secretariat for the Board.

Following the example of the European Data Protection Board, the AI Board would be tasked with ensuring consistent application of the AI rulebook across the bloc while also advising on secondary legislation, codes of conduct, and technical standards.

The co-rapporteurs suggest adding to the Office’s tasks, assisting national authorities in establishing regulatory sandboxes and organising periodic consultations with the advisory forum.

Scientific panel

Setting up a scientific panel of independent experts also made its way into the text. The Commission is to appoint the experts based on their expertise, independence from AI companies and accuracy of their work.

For the co-rapporteurs, the board composition should ensure gender and geographical representation.

The panel advises the AI Office about General Purpose AI (GPAI) models, in particular, contributing to developing methodologies for evaluating the models’ capabilities and advising on classifying models with systemic risks.

Moreover, the panel experts should alert the AI Office when they consider that a GPAI model presents a concrete risk at the EU level or meets the requirements to classify it as a model with systemic risks.

The leading MEPs also suggest that the scientific experts advise on the relevant benchmarks for GPAI models and the distinction between the GPAI model and system categories.

Concerning national authority, the experts would support their work and cross-border investigations upon request. The AI office would have to manage potential conflicts of interest for the panel members.

Advisory forum

The Commission’s text picks up the European Parliament’s idea to set up an advisory forum for stakeholders like industry representatives, SMEs, start-ups, civil society and academia to provide technical expertise to the Board and AI Office.

[Edited by Alice Taylor]