New compromise amendments for EU AI Act high-risk areas in attempt to end negotiations

Negotiations on the upcoming EU AI Act could be drawing to a close as the European Parliament’s co-rapporteurs for the Act propose compromise amendments, primarily to the list of high-risk AI application. A compromise conformity assessment procedure for AI providers has also been shared.

The co-rapporteurs, MEPs Brando Benifei and Dragoș Tudorache, extended the list of high-risk areas of use, according to a compromise version of the Act’s Annex III seen by Euractiv. The concept of biometric identification and categorization would be extended to include uses such as Lensa, the avatar-generation app based on facial features.

They have added live and ex-post remote biometric identification in private spaces to the list. For public spaces it is limited to retrospective identification, though use cases include remote biometric categorization and emotion recognition systems in publicly-accessible spaces, reports Euractiv.

Safety components for road, rail and road traffic have been included in the high-risk list. Systems affecting work and employment such as automated decision-making for termination have been added.

Many other use cases were amended or reworded. Further definitions have been added such as for biometric identification and categorization. The ban on AI in predictive policing was maintained.

Conformity assessments

The co-rapporteurs recommended that the president of the European Commission be able to amend the process of conformity assessments, but only after consulting with the EU’s Artificial Intelligence Office and relevant stakeholders.

And, looking out for smaller vendors, the pair suggested that audit fees to categorize high-risk code be made proportional to company size and market share.

They also recommended that a market surveillance authority would need to get a judge’s authorization to deviate from conformity assessment procedures. Other member states and the European Commission would have to be notified of the request and the approval.

After a December compromise on the Act, world-first legislation that seeks to protect people’s safety and rights and establish a clear regulatory environment for the development of trustworthy AI, the organs of the EU are attempting to reach a final agreement with the aim of adopting the act by the end of 2023.

The co-rapporteurs proposed a tenth set of amendments in mid-December that added biometric data to high-risk criteria. Following the compromise of Europe’s digital ministers, Germany appeared to pull back from other EU members and side with the European Parliament, which has taken a much stricter stance on technologies such as remote biometric identification, wanting to ban both live and ex-post usage.

Rights groups criticized the December compromise, claiming it watered down the Act. EDRi produced a guide for technically and legally defining remote biometric identification, called for its total ban and suggested rewording for parts of the draft act.

With files from Jim Nash.

Article Topics

AI  |  AI Act  |  biometric identification  |  biometrics  |  EU  |  facial recognition  |  legislation  |  surveillance