According to the Head of Office of Rapporteur Axel Voss (EPP, Germany), a provisional agreement on the AI Act proposal by the end of this year may be possible, but it depends on the progress made during upcoming discussions.
The fourth trilogue on the AI Act is scheduled for 25 October, instead of 26 October. A provisional agreement being reached by the end of this year (2023) is possible, but still highly uncertain.
Agreements have been reached on the following issues:
(i) obligations of providers of high-risk AI systems and other parties (Articles 16-27, excluding Articles 23a and 24);
(ii) notifying authorities and notified bodies (Articles 30-39);
(iii) standards, conformity assessments, certificates, and registration (Articles 40-50, excluding Articles 41 and 47);
(iv) regulatory sandboxes (Articles 53-55, with the exception of real world testing);
(v) requirements for high-risk AI systems (Articles 8-12, excluding Annex IV);
(vi) registration to be submitted upon the registration (Articles 51, 60, and Annex VIII, excluding digital gatekeepers);
(vii) market surveillance, post-market monitoring, and information sharing (Articles 61-68, excluding ‘access to source code’ and parts of Article 68a);
(viii) confidentiality, penalties, and delegation of powers (Articles 70-74); and
(x) final provisions (Articles 75-85, excluding Article 82b and parts of Articles 83-85).
A final agreement could not be reached concerning the classification of high-risk AI (Article 6), suggesting that AI systems falling within Annex III (for creditworthiness assessment and credit scoring) failing to being considered high-risk unless there are exceptional circumstances.
FEBIS will follow the developments closely, exchanging with stakeholder on the progress.
Source: EU News/FEBIS