Legislative Decree No 179 of 7 October 2026 was published in Official Gazette No 234 of 8 October 2026 and enters into force on 23 October 2026 (Official Gazette (Gazzetta Ufficiale)). It aligns the Italian legal order with Regulation (EU) 2024/1689, the AI Act (EUR-Lex), “as regards the powers of the national Authorities and the use of artificial intelligence in training” (our translation). When we examined the Italian implementation of the AI Act, between supervisory authorities, penalties and the prohibition on employment decisions, the analysis was based on press release No 185 of the Council of Ministers of 4 August 2026. The published text calls for some updates and shifts the focus to internal organisation.
From the press release to the text: European ceilings and no reference to the Digital Omnibus
The press release announced ceilings lower than the European ones and coordination with Regulation (EU) 2026/1744, the Digital Omnibus on AI, which would have tied the entry into force of the penalties to that of the corresponding obligations and prohibitions (Italian Government). The published text confirms neither the reduced ceilings nor the coordination with the Digital Omnibus. Article 23 takes over the amounts in Article 99 of the AI Act (AI Act Service Desk): EUR 35 million or 7 per cent of total worldwide annual turnover for prohibited practices, EUR 15 million or 3 per cent for the obligations of providers, deployers and other operators, EUR 7.5 million or 1 per cent for incorrect information. On one point, however, the text goes further than the press release: Article 41 expressly includes “performance evaluations” among final decisions.
The national authorities: notification to AgID, supervision to ACN and sectoral competences
The Agency for Digital Italy (AgID) is the notifying authority for the assessment, designation, notification and monitoring of conformity assessment bodies (Article 4) and may rely on the national accreditation body (Article 13(5)). The National Cybersecurity Agency (ACN) is the market surveillance authority and the single point of contact (Articles 5 and 6). The Bank of Italy, CONSOB, the companies and stock exchange commission, and IVASS, the insurance supervisory authority, supervise the high-risk systems of their respective financial and insurance services; the Italian Data Protection Authority (Garante per la protezione dei dati personali) acts “only to the extent provided for in Article 74(8) of the AI Regulation” (AI Act Service Desk). Article 5, however, preserves the competences of the European AI Office under Articles 75, 88 and 89 of the AI Act (Official Gazette (Gazzetta Ufficiale)), and after the Digital Omnibus Article 75 reserves to that Office, subject to exceptions, the systems based on general-purpose models of the same provider or the same undertaking and those integrated into very large online platforms and search engines (EUR-Lex): as Nicola Fabiano observes, not every system therefore falls under an Italian authority (NicFab). For a business, the first check concerns which authority will supervise it, and the answer depends on the sector and the type of system. Article 14 also entrusts ACN, implementing Article 49(5) of the AI Act, with the national registration of the systems in point 2 of Annex III, on critical infrastructure (AI Act Service Desk).
The penalty regime: European ceilings, an additional national band and alternative measures
Article 23(4) introduces a band that Article 99 does not provide for: failure to comply with Articles 27 and 86 of the AI Act is punishable “up to EUR 1 000 000 or, if higher, up to 0.5 per cent of total worldwide annual turnover for the preceding financial year”. The fundamental rights impact assessment and the right to explanation now carry their own national penalty. The former is borne, among others, by deployers that are bodies governed by public law or private entities providing public services (AI Act Service Desk); on its relationship with the DPIA we refer to ISO/IEC 42005 as a common method between FRIA and DPIA. The latter belongs to anyone subject to a decision taken by the deployer on the basis of a system listed in Annex III, excluding point 2, which produces legal effects or similarly significantly affects that person in a way that, in their view, has an adverse impact on their health, safety or fundamental rights (AI Act Service Desk); under these conditions it also covers decisions based on the systems in point 4, concerning employment and workers’ management (AI Act Service Desk).
For infringements of minor seriousness Article 22 allows, as an alternative, an order to eliminate the infringement within a time limit or, once the infringement has ceased, a public statement on the infringement and the person responsible; failure to comply with the order increases the penalties “by up to one third”. The penalty decision “is published without delay and in summary form” on the authority’s website (Article 24(4)). Even the public statement, an alternative to the financial penalty, exposes the business to reputational harm.
Decisions on the employment relationship and safety: human intervention, reasons and risk assessment
Article 41, “without prejudice to Article 22 of Regulation (EU) 2016/679”, requires the employer to ensure that final decisions on the establishment, modification or termination of the relationship are not taken solely on the basis of automated processing, reserving the decision to a natural person “who exercises effective and autonomous power”. The screening of applications remains excluded; in our view the reference to Article 22 preserves the safeguards of the GDPR in that phase, the same safeguards on which the fine imposed on Uber for the automated deactivation of drivers is based. The worker’s consent does not make lawful a decision taken solely by automated means. The worker is also entitled to “an intelligible statement of the reasons for the decision concerning them”, indicating any influence of the systems and the main parameters considered (paragraph 3). Dismissals and other decisions adopted in breach are null and void (paragraph 4): the risk for the employer is not only the penalty, but the nullity of the act.
Under Article 42, systems that affect the organisation of work, production rates, the manner in which work is performed or decisions relevant to safety are assessed “as part of the risk assessment referred to in Article 28 of Legislative Decree No 81 of 2008”, and workers receive information and training on the specific risks; on the consequences for the risk assessment document and for the prevention figures, see our analysis of artificial intelligence and safety at work.
The internal roles to be covered: AI governance, cybersecurity and product
The provisions examined identify the obligated parties: employer, provider and deployer. They do not assign tasks to specific internal roles, and the allocation remains a choice for the business. The correspondences that follow are our own assessment, each with the provision that supports it and its limit.
The first area to be covered is overall governance: inventory of systems, competent authority for each, reporting to the board. In our view it falls to the Chief AI Officer, the role that answers for artificial intelligence before top management, to whom we have assigned risk governance and reporting. The link with the decree runs through Article 23(6), which refers to the criteria in Article 99(7) of the AI Act, including the degree of cooperation with the authorities and the degree of responsibility of the operator “taking into account the technical and organisational measures implemented by it” (AI Act Service Desk). The criterion looks at measures, not roles; we consider, however, that a documented allocation of these tasks helps to demonstrate them.
The second is the security of high-risk systems. Article 15 of the AI Act requires them to be “resilient against attempts by unauthorised third parties to alter their use, outputs or performance by exploiting system vulnerabilities”, with measures against data poisoning and model poisoning, adversarial examples and confidentiality attacks (AI Act Service Desk). Article 16, point (a), requires the provider to ensure compliance with the requirements of Section 2 of Chapter III, which includes Article 15 (AI Act Service Desk), and Article 16 falls within the band of up to EUR 15 million or 3 per cent in Article 23(2) of the decree. Those techniques correspond to a large extent to the attacks described with regard to the AI Security Specialist, the profile called upon to stop the attacks specific to AI systems: we therefore consider that the technical safeguarding of Article 15(5) falls to this role, without prejudice to the provider’s responsibility.
The third is the product life cycle. Article 13 requires the operation of the system to be “sufficiently transparent to enable deployers to interpret a system’s output and use it appropriately” and the instructions for use to indicate, where applicable, the technical characteristics relevant to explaining its output (AI Act Service Desk). In our view the explanation under Article 86 and the statement of reasons under Article 41(3) will be all the easier the more the provider has built such information into the design: a task we attribute to the AI Product Manager, who governs the life cycle of AI-based products. The limit is evident: this role operates at the provider, whereas the statement of reasons and the explanation remain obligations of the employer and the deployer.
Critical aspects: uncoordinated deadlines and implementing acts still to be adopted
The first aspect is the lack of coordination with the Digital Omnibus, which set 2 December 2027 for the application of Sections 1, 2 and 3 of Chapter III to Annex III systems and 2 August 2028 for Annex I systems (EUR-Lex). The penalties apply from 23 October 2026, and Articles 16, 26 and 27 are in Section 3 of Chapter III. For high-risk systems, the national penalties enter into force before the European obligations they are meant to enforce. We consider that no penalty can be imposed for an obligation that is not yet applicable, because Article 23 punishes failure to comply with obligations under the AI Act that cannot be enforced until those dates; the decree, however, contains no clause on when the penalties take effect (see NicFab). For Article 86, located in Chapter IX but referring to systems whose requirements are deferred, the point from which the penalty becomes enforceable remains to be clarified.
The second concerns the acts to be adopted: AgID must define the procedures for notifying bodies (Articles 4 and 13), the arrangements for registration with ACN are left to a decree of the Agency’s Director General (Article 14(2)) and those for exercising supervisory powers to a regulation adopted by decree of the President of the Council of Ministers (Article 15(3)); the professional orders have six months to adapt their training regulations (Article 47(5)).
Conclusions
The employer must reserve final decisions on the relationship to a natural person, give reasons for them on request and take artificial intelligence into account in the risk assessment; providers and deployers retain the obligations of the AI Act according to the European calendar, now backed by national penalties with full ceilings. What is missing are the notification procedures and a clarification of the coordination with the Digital Omnibus. The time between 23 October 2026 and 2 December 2027 should be used to assign roles and document procedures: an inventory of systems with the competent authority, a documented procedure for personnel decisions, an updated risk assessment and the formal allocation of the three areas of responsibility.
Author: Valentina Grazia Sapuppo
