Legislative Decree No 179 of 7 October 2026, published in Official Gazette (Gazzetta Ufficiale) No 234 of 8 October 2026 and in force from 23 October, devotes Article 42 to «Artificial intelligence, health and safety in the workplace» (our translation) (Gazzetta Ufficiale). The provision does not create a separate obligation: it places artificial intelligence systems within the risk assessment required by Legislative Decree No 81 of 9 April 2008, the Consolidated Act on health and safety at work. The consequences for risk assessment documents, site plans and training are left largely to the interpreter; Article 41, on decisions concerning the employment relationship, is a distinct subject, which we considered when discussing Italy’s implementation of the AI Act and the ban on employment decisions.
Article 42 of Legislative Decree 179/2026: when AI becomes a risk to be assessed
The Article has four paragraphs. The first and the fourth entrust the Ministry of Labour and Social Policies with coordinating protective measures and with promoting artificial intelligence «also as a tool for preventing accidents at work and occupational diseases», in liaison with the National Labour Inspectorate (Ispettorato nazionale del lavoro). The second and the third concern the employer.
Paragraph 2 provides that the use of systems affecting «the organisation of work, production rhythms, the manner in which work is performed or decision-making processes relevant to safety» is assessed «within the risk assessment referred to in Article 28» of the Consolidated Act. The criterion is the effect on work, not the technology as such. A system that allocates tasks on the basis of recorded times or a camera that flags dangerous behaviour falls within the rule for what it does to the organisation, not because it uses machine learning.
Article 28 already required the assessment of «all risks to the safety and health of workers», including those connected with work-related stress (Normattiva), and the general protective measures of Article 15 include ergonomic principles aimed at reducing the effects «of monotonous and repetitive work» (Normattiva). We therefore consider that paragraph 2 does not extend the subject matter of the assessment, but makes one of its contents explicit, which remains the non-delegable responsibility of the employer (Article 17).
The European Agency for Safety and Health at Work (EU-OSHA) notes that worker management through artificial intelligence involves opportunities and risks, including reduced autonomy and other psychosocial risks, resulting in stress and fatigue (EU-OSHA). A pace set by an algorithm is a risk factor to be assessed just like a pace set by a production line.
Updating the risk assessment and construction site safety plans: timing and conditions under Legislative Decree 81/2008
Article 42 does not say when the document must be updated. As Andrea Dari observes, the decree does not link every use of AI to an automatic update, and the need to review the assessment must be verified against the conditions of Article 29 (cf. Ingenio). The cases listed in Article 29(3) include «changes to the production process or to the organisation of work that are significant for the health and safety of workers»: the assessment must be reviewed within thirty days, with «immediate notification» to the workers’ safety representative (Normattiva).
In our view the four areas of Article 42 operate as an indicator of significance: a system that alters work rates, the allocation of tasks or safety decisions constitutes, as a rule, an organisational change relevant under Article 29. Tools that do not touch the performance of work remain outside.
On temporary or mobile construction sites the locus is different. The operational safety plan is drawn up by the employer of the contractor for the individual site «pursuant to Article 17(1)(a)» (Normattiva), and Article 96(2) provides that acceptance of the safety and coordination plan and drafting of the operational safety plan constitute, for that site, compliance also with Article 29(3) (Normattiva). We therefore consider that vision systems signalling access to dangerous areas, or software scheduling works and interferences between them, must be assessed in the operational safety plan of the undertaking that uses them. Where they are envisaged by the client, the place for coordination is the safety and coordination plan, with requirements «related to the complexity of the work to be carried out and to any critical phases of the construction process» (Normattiva). The coordinator for the execution phase adapts that plan «in relation to the progress of the works and to any changes that have occurred» and verifies the adaptation of the operational plans (Article 92(1)(b)), including where a system is introduced once works are under way.
RSPP, occupational physician and RLS: the role of the prevention figures
Article 42 does not mention the head of the prevention and protection service (RSPP), the occupational physician (medico competente) or the workers’ safety representative (RLS). Their involvement derives from Article 29: the employer carries out the assessment with the RSPP and, where health surveillance applies, with the occupational physician, after consulting the RLS. The occupational physician collaborates in the assessment and in the measures protecting «the psycho-physical integrity of workers» (Normattiva), a perimeter within which we consider that work rates and controls imposed by a system fall. The RLS is consulted «in advance and in good time» on the assessment and receives the related documentation, including on the organisation of work (Normattiva); in our view, the RLS’s specific training on the risks in the areas represented (Article 37(10)) should by now cover the systems present in the company.
Paragraph 3 of Article 42 requires «adequate information and training of workers on the specific risks connected with the use of artificial intelligence systems, and on the prevention and protection measures adopted». The wording echoes Articles 36 and 37, and Article 37(4)(c) already requires training and, where provided, instruction on the occasion of «the introduction of new work equipment or new technologies» (Normattiva). This is not a generic course on artificial intelligence, but training on the risks of the system the worker uses or is subject to: the limits of reliability of its alerts, how to act when the system errs or stops, who may correct its instructions.
This channel must be kept distinct from the information required by Article 1-bis of Legislative Decree No 152 of 26 May 1997 on fully automated decision-making or monitoring systems, referred to in Article 41(3) of the decree and also communicated to trade union representatives (Gazzetta Ufficiale). Workers must know how the system works and, separately, how to protect themselves from its effects. We consider it advisable to coordinate the two flows.
The AI Act and the Machinery Regulation: worker management and AI safety components
Annex III, point 4(b), of the AI Act classifies as high-risk AI systems intended, among other things, «to allocate tasks based on individual behaviour or personal traits or characteristics or to monitor and evaluate the performance and behaviour of persons in such relationships» (AI Act Service Desk): to a large extent, the same systems that affect organisation and work rates. Article 26 requires the deployer to assign human oversight to persons who have «the necessary competence, training and authority» and, where it is an employer, to inform workers’ representatives and the affected workers before putting the system into service, obligations applicable to Annex III systems from 2 December 2027; it also states that these obligations «are without prejudice to other deployer obligations under Union or national law» (AI Act Service Desk). The Consolidated Act therefore continues to apply alongside the AI Act. We consider that the person overseeing the system should be identified in the risk assessment document among the roles entrusted to «persons with adequate skills and powers» (Article 28(2)(d)).
On machinery the link is textual. Regulation (EU) 2026/1744 moved Regulation (EU) 2023/1230 from Section A to Section B of Annex I to the AI Act and required the Commission to add, by means of delegated acts, requirements to Annex III of the Machinery Regulation for high-risk AI systems that are its safety components, to apply by 2 August 2028 (EUR-Lex), as we saw for self-learning machines under Regulation 2023/1230. Among the ergonomic principles the manufacturer must take into account, that Regulation lists «avoiding a machine-determined work rate» (EUR-Lex). The work rate is thus both a design requirement and a subject of the employer’s assessment.
In the background lies Article 11 of Law No 132 of 23 September 2025, under which artificial intelligence is used «to improve working conditions, to protect the psycho-physical integrity of workers» (Gazzetta Ufficiale): alongside the reservation of human decision-making in Law 132/2025, Article 42 gives that principle a familiar operational instrument.
Critical aspects: criteria still to be defined and European timelines out of step
Paragraph 2 does not define when a system «affects» the four areas, nor does it refer to guidance documents, and paragraphs 1 and 4 assign tasks to the Ministry without deadlines or instruments. Article 29(6-quater) already provides for a ministerial decree on tools supporting the assessment, including computerised tools based on the European OiRA prototype: we consider this the natural place for guidance on AI systems.
The penalties in Article 23 of the decree concern obligations under the European Regulation, including those of the deployer under Article 26, not Article 42. Those of the Consolidated Act remain: Article 55 punishes the adoption of the document without following the procedures of Article 29(2) and (3), and the breach of the information and training obligations of Articles 36 and 37 (Normattiva). The obligation to train on AI risks is enforced only to the extent that it coincides with those already laid down in the Consolidated Act.
Finally, the calendar. The assessment is due from 23 October 2026, while the obligations of providers of Annex III systems, including the instructions for use, apply from 2 December 2027 (EUR-Lex). For over a year employers will have to assess systems whose limits and performance the provider is not yet required to describe under the AI Act. We consider that the gap should be filled contractually, by requesting from the provider the information needed for the assessment.
Conclusions
Article 42 leaves the employer, with the RSPP, the occupational physician and after consulting the RLS, with the task of identifying the relevant systems, assessing them in the risk assessment document or, on construction sites, in the operational safety plans, and training workers on the specific risks. Ministerial criteria are missing, as is the documentation that the AI Act will make mandatory only from December 2027. In the meantime, it is advisable to map the systems in use according to the four areas of the provision, check the conditions for review under Article 29, update site plans and plan training on the concrete system, not on artificial intelligence in abstracto.
Author: Valentina Grazia Sapuppo
