Italy has become the first country in the European Union to write artificial intelligence directly into its workplace safety rulebook, with a sweeping decree that took effect on October 23, 2026, requiring employers to assess AI systems in their mandatory risk documents and barring hiring, firing or disciplinary decisions made by an algorithm alone. The measure, Legislative Decree No. 179 of October 7, 2026, published in the Gazzetta Ufficiale No. 234 on October 8, adapts Italian law to the EU AI Act and runs to 54 articles covering national AI authorities, training, and the use of AI at work.
Illustration: Under Italy's new decree, employers must assess AI systems in the same risk document that covers other workplace hazards.
What Happened
From October 23, 2026, every employer with staff in Italy, public or private, large or small, must treat AI as an occupational health and safety matter. The decree entered into force with no transition period, meaning the obligations applied immediately, a point that has caught the attention of employment lawyers and multinational companies operating in Italy.
The best-known provision sits in Article 41 of the decree, in the chapter on AI in training, the professions, work, health and public administration. According to consistent reporting by Italian employment-law practitioners, Article 41 provides that decisions on the creation, modification or termination of an employment relationship, including disciplinary measures and performance evaluations, may not rest solely on automated processing. The final decision is reserved to a natural person exercising effective and autonomous decision-making power, and a worker's consent does not lift the rule. Decisions or dismissals taken in breach are described as void. Workers may also demand an intelligible explanation of a decision, including the role the AI system played and the main parameters it considered.
A second, less headline-grabbing provision may matter just as much in practice. Article 42 amends Italy's Consolidated Safety Act, Legislative Decree 81/2008, the Testo Unico on workplace health and safety. That statute already requires every employer to compile a Documento di Valutazione dei Rischi, the DVR, a risk assessment document identifying workplace hazards and the prevention and protection measures in place. From October 23, any AI system that meets the decree's functional workplace-effect test must be assessed within that same DVR, with the employer documenting the specific risks, the prevention and protection measures, and the training and information programme provided to affected workers.
Background
Italy is the first EU member state to embed an AI safety obligation directly inside its national occupational safety and health documentation framework, according to reporting on the decree. By placing AI inside the DVR, compliance becomes a routine labour inspection matter, enforceable through existing domestic inspection mechanisms, rather than a specialised technology regulation waiting for EU-level harmonised standards or sector guidance.
Oversight is shared. Italy's Ministry of Labour coordinates safety measures when AI enters the workplace, while the National Labour Inspectorate, the Ispettorato nazionale del lavoro, holds existing inspection and enforcement powers over DVR compliance. The decree also names Italy's national AI authorities: AgID as the notifying authority and ACN, the national cybersecurity agency, for market surveillance and as the single point of contact, with Banca d'Italia, CONSOB and IVASS covering high-risk systems in their financial sectors. Penalties under the framework can reach 35 million euros or 7 percent of turnover, according to summaries of the decree.
The decree's four areas for DVR review show how broad the net is. Employers must examine the organisation of work and tasks; operational processes, including personal protective equipment fitted with sensors; control over work rhythms where shifts are scheduled algorithmically; and processes and maintenance carried out with the help of predictive algorithms. Specific strains arising from digitalisation sit at the centre of the assessment: technostress, anxiety about algorithmic monitoring, and potential errors stemming from faulty model outputs, so-called AI hallucinations, all have to be recorded and evaluated systematically.
Illustration: A human must make the final call on hiring, firing and discipline — an algorithm alone is not enough under the new rules.
The decree also sits within a wider European timetable. On August 2, 2026, the EU AI Act's Article 50 transparency obligations became directly applicable, and the European Commission's AI Office gained powers to investigate and fine providers of general-purpose AI models. Italy's decree is the national transposition that turns those European rules into day-to-day Italian workplace obligations, and it arrives as other countries debate their own approaches. In the United States, for example, Senator Maria Cantwell unveiled a framework on October 7, 2026, for mandatory frontier AI safety standards, including independent testing and liability for harm, in contrast to the White House's voluntary principles, while Anthropic has separately pitched Congress its own draft safety bill. Italy, by contrast, has chosen a route that runs through the factory floor and the office, not just the model lab.
Why It Matters
For workers, the decree changes who, or what, can decide their working lives. Algorithmic management already shapes hiring, scheduling, performance scoring and discipline in warehouses, call centres, delivery platforms and offices. By requiring a human with real, autonomous decision-making power, and by making algorithm-only dismissals void, the law gives workers a concrete remedy rather than a vague principle. The right to an intelligible explanation, delivered through a person, also matters: a worker can now ask what the machine counted, and how much weight it carried.
For employers, the DVR obligation turns AI governance into paperwork with teeth. Companies must now inventory every tool that scores, ranks, monitors or recommends action on employees, including features switched on inside standard HR software suites, and assess each one for risks to workers' psychophysical wellbeing. Multinationals running global HR tools from outside Italy are not exempt for their Italian workforce, according to employment-law reporting. The absence of a transition period means that inventory and assessment work is already overdue for many firms.
For the rest of Europe, Italy is a test case. Other member states transposing the EU AI Act will watch whether labour inspectors, rather than technology regulators, prove effective enforcers, and whether the DVR model spreads. If it works, the humble risk assessment document, a staple of European workplaces for decades, could become the main vehicle for AI accountability at work across the continent.
Illustration: Sensor-equipped protective gear and predictive algorithms are among the systems employers must now assess for worker safety.
What Happens Next
The immediate next step is compliance work. Employment lawyers advise firms with staff in Italy to list every AI tool affecting workers, name the human decision-maker for each decision type covered by Article 41, check that worker information notices describe the systems actually in use, prepare a process for giving human explanations on request, and review workplace risk assessments wherever software controls workloads or pace. Labour inspectors can now test those documents during routine visits.
One controversy is already brewing. Candidate search and screening are not treated as final decisions under the decree, even where screening excludes someone from later stages, according to the official text as reported. Italian data protection lawyer Guido Scorza criticised that carve-out in Il Fatto Quotidiano on October 23, 2026, reportedly calling it a serious error of assessment that threatens to deprive qualified people of jobs because of bias in screening algorithms, and citing Amazon's abandonment of an early algorithmic recruiting project in 2015 after it showed a preference for men over women. Critics argue that if the first filter is biased, a human final decision later cannot undo the harm. Supporters of the decree reply that screening remains covered by data protection law, including GDPR Article 22 on fully automated decisions. Expect that debate, and possibly court challenges, to shape how the law is interpreted.
More broadly, watch for guidance and enforcement practice. The decree encourages cross-border collaboration with the regions and in sectors such as defence, justice and fintech, and a regulatory sandbox is provided for. How strictly inspectors enforce DVR completeness, and whether fines follow, will determine whether Italy's experiment becomes a model for Europe or a cautionary tale about ambitious rules meeting messy reality. Either way, the principle is now law in one of Europe's largest economies: when AI touches your job, a person, not just a program, must answer for the decision.
Sponsored: explore today's featured sponsored offers
FAQ
What is Italy's Decree 179/2026?
Legislative Decree No. 179 of October 7, 2026, published in the Gazzetta Ufficiale No. 234 on October 8, 2026, is Italy's 54-article law adapting national law to the EU AI Act. It covers national AI authorities, training, and the use of AI at work, and entered into force on October 23, 2026, with no transition period.
Can an AI system fire or discipline a worker in Italy now?
No, not on its own. Under Article 41, decisions on hiring, changing or ending employment, including discipline and performance evaluations, may not rest solely on automated processing. The final decision must be made by a natural person with effective, autonomous power, and dismissals in breach are described as void. A worker's consent does not lift the rule.
What must employers put in their risk assessment document?
Employers must assess AI systems within the DVR, the Documento di Valutazione dei Rischi required by Legislative Decree 81/2008, documenting the risks, prevention and protection measures, and worker training. The assessment covers work organisation, sensor-equipped protective equipment, algorithmic scheduling and work rhythms, and predictive maintenance, including technostress and AI errors.
Who enforces the new rules?
Italy's Ministry of Labour coordinates AI workplace safety measures, and the National Labour Inspectorate enforces DVR compliance through existing inspection powers. AgID is the notifying authority and ACN handles market surveillance, with Banca d'Italia, CONSOB and IVASS covering their financial sectors. Penalties can reach 35 million euros or 7 percent of turnover.
Does the decree cover AI screening of job candidates?
Not as a final decision: candidate search and screening are not treated as final employment decisions even where they exclude applicants, according to the official text as reported. Critics, including lawyer Guido Scorza, argue this leaves a gap, while supporters note that data protection law, including GDPR Article 22, still applies to automated screening.
Sources
- Legislative Decree No. 179 of October 7, 2026, published in the Gazzetta Ufficiale No. 234, October 8, 2026, as summarised by Noze, October 8, 2026, and Remote Work Europe.
- TechTimes, "Italy's AI Workplace Safety Mandate Takes Effect Thursday: Scheduling Tools Covered," October 22, 2026.
- Ad-hoc-news, "Italy Tells Employers: Put AI on the Risk Register by 23 October," reporting umbria.it, October 21, 2026.
- PwC Italia Blog, October 22, 2026, on Article 41, and Il Fatto Quotidiano, October 23, 2026, on the screening carve-out debate, as reported in October 2026 coverage.
- PYMNTS, "Push for Mandatory AI Safety Rules Gains Momentum in Washington," October 8, 2026, on Senator Cantwell's framework.
Comments
Post a Comment