AI governance
Projet de loi portant diverses dispositions d'adaptation au droit de l'Union européenne (DDADUE), Articles 24 to 24 terdecies, Authorities Competent Under the AI Act
Articles 24 à 24 terdecies texte adopté n° 63 (2025-2026), Sénat, 18 février 2026 (autorités compétentes, règlement sur l'intelligence artificielle)Texte adopté n° 63 (2025-2026), adopted by the Sénat in first reading on , Sénat
Before the second chamber, dated , as of . Binds public and private bodies.
What this law does
Article 24 of the bill, as adopted by the Sénat, would insert a title V bis on artificial intelligence systems into the loi n° 2004-575 du 21 juin 2004, whose Article 55-1 would subject AI systems to the AI Act within the conditions and limits that Regulation defines.
Proposed Article 55-2 would let the authorities competent under Article 70 of the AI Act share the information and documents they hold, and would require them to inform one another before bringing proceedings or imposing a sanction for the same facts. Proposed Article 55-3 would let the market surveillance authorities and their agents use the technical expertise of the State administrative service named in Article 36 of the loi n° 2021-1382 when checking AI systems against the AI Act.
Proposed Article 55-5 would authorize agents of the direction générale des entreprises to find and record breaches of Section 4 of Chapter III of the AI Act by notified bodies in the toy sector.
Article 24 bis would authorize the Government to legislate by ordinance, within six months of promulgation, to adapt the overseas codes and laws to the AI Act and to define the access conditions, safeguards and authorization procedures for the regulatory sandboxes and real-world testing that the AI Act provides.
Article 24 quinquies would add to the list in Section L. 511-7 of the Code de la consommation Article 5(1)(a), (b) and (c) and Article 50(1), (2) and the first subparagraph of (4) of the AI Act, and the requirements for high-risk systems under point 3 of Annex III in the field of vocational training. Article 24 duodecies would make the Commission nationale de l’informatique et des libertés one of the authorities competent under Article 70 of the AI Act.
Proposed Article 20-12 of the loi n° 86-1067 would have the Autorité de régulation de la communication audiovisuelle et numérique supervise compliance with Chapter II of the AI Act for the prohibited practices in Article 5(1)(a) and (b), and with Chapter IV for the transparency duties in Article 50(1), (2) and (4).
Article 24 undecies would add to the supervisory missions in Section L. 612-1 II of the Code monétaire et financier the monitoring of the AI Act requirements for high-risk systems under point 5(b) and (c) of Annex III that supervised financial entities place on the market, put into service or use.
Article 24 nonies would add the market surveillance of high-risk AI systems that are safety components of medical devices, or are themselves medical devices, to the competences listed in Section L. 5211-2 of the Code de la santé publique.
Proposed Article L. 116-1 of the Code de justice administrative would make the Conseil d’État responsible for market surveillance of the high-risk AI systems that the administrative courts put into service or use in their judicial function, in place of the Commission nationale de l’informatique et des libertés. Proposed Article L. 1340-2 of the Code de la défense would have a decree of the Conseil d’État designate the market surveillance authority for each domain of critical infrastructure in which high-risk AI systems serve as safety components.
Proposed Article L. 1340-4 of the Code de la défense would let that authority's agents, after giving the operator at least ten working days to comment, order a high-risk AI system's non-conformity to be ended, its marketing or use to be suspended or banned, its recall or withdrawal, or warnings to be published.
Proposed Article L. 1340-5 of the Code de la défense would have that authority impose the administrative fines in Article 99 of the AI Act on operators of high-risk AI systems for the breaches its agents record.
What it requires