Italy’s New AI Decree Bans Fully Automated Decisions on Workers

Employers in Italy that use artificial intelligence must now leave the final word on hiring, discipline, performance reviews and dismissal to a person. Legislative Decree 179/2026, published in the Gazzetta Ufficiale on 8 October, enters into force on 23 October 2026.

The decree adapts Italian law to the EU AI Act, the bloc’s main AI regulation. Its headline purpose is to set out the powers of national authorities and how AI is used in education and training. But its Article 41 matters most to employers. It says final decisions on a worker’s employment cannot be taken “solely on the basis of automated processing” (translated), and any dismissal that breaks the rule is null. Professional services firm PwC Italia calls it “the main change for employers”.

Article 41 reserves the final decision for a person

According to the official text, an employer using AI systems must ensure that final decisions on starting, changing or ending an employment relationship, “including disciplinary measures and performance evaluations”, are not taken solely through automated processing (translated). “The definitive decision is in any case reserved to a natural person, who exercises effective and autonomous power,” the article says (translated).

The article applies without prejudice to Article 22 of the EU’s General Data Protection Regulation (GDPR), which already limits decisions based solely on automated processing. Two further lines narrow and harden the rule. First, searching for and selecting candidates does not count as a final hiring decision, “even when they result in non-admission to the subsequent stages of selection” (translated). Second, “The worker’s consent does not constitute an exception to the prohibition” (translated).

The use of these systems must also respect workers’ dignity and privacy and the principle of non-discrimination.

Workers can ask a person to explain the decision

Before any processing begins, employers must give the information required by Article 1-bis of Legislative Decree 152/1997. Workers are entitled, on request and “through the intervention of a natural person”, to an intelligible explanation of a decision that concerns them, including any influence AI had on it and the main parameters considered (translated).

That information duty comes from Italy’s framework AI law. Article 11 of Law 132/2025, in force since 10 October 2025, says AI at work must be “safe, reliable, transparent” and cannot run counter to human dignity (translated). It requires employers to tell workers when AI is used, under Article 1-bis. That provision lists what employers must disclose about automated systems, from their purpose and logic to the main parameters and the accuracy and cybersecurity levels. Workers or their unions can request the data and must get a written answer within 30 days. Systems protected by industrial and commercial secrecy are exempt from the information duties.

A dismissal in breach is null

Article 41’s sanction is direct. “A dismissal ordered in breach of paragraph 1 is null. Decisions taken by the employer in breach of paragraph 1 are null,” it reads (translated).

Article 42 adds health and safety duties. AI systems that affect how work is organised, the pace of production, how tasks are carried out or safety-related decisions must be covered in the risk assessment required under Legislative Decree 81/2008. Employers must also inform and train workers on the specific risks.

AgID and ACN lead enforcement of the wider AI Act

The rest of the decree sets out who enforces the AI Act in Italy. The Agency for Digital Italy (AgID) is the national notifying authority. The National Cybersecurity Agency (ACN) is a market surveillance authority for AI systems, with the Bank of Italy, CONSOB and IVASS covering high-risk AI in lending, financial services and insurance. Fines for using AI practices banned by the AI Act reach €35 million or 7% of worldwide annual turnover, whichever is higher. Breaches of other listed obligations, including those on deployers, carry up to €15 million or 3%.

TSN has followed the AI Act’s practical rollout, including its first harmonised standard, and workplace AI rules elsewhere, from California’s 13 AI bills to a South Korean bill on AI-driven layoffs.

Consent cannot rescue a fully automated decision

For HR teams, the consent clause is the line to read twice. Many compliance programmes treat a signed acknowledgement as a safe harbour. Article 41 removes that option for final employment decisions. The remedy compounds the risk. A fine is a cost, but a null dismissal unwinds the decision itself. The practical test will be the phrase “effective and autonomous power”. A manager who clicks to approve an algorithm’s ranking without reviewing it may not meet that standard, and PwC reads the rule the same way: the person must assess the output “rather than simply accepting it”.

The documentation burden shifts too. Because a worker can ask for an intelligible explanation, given through a person and covering any influence AI had and the main parameters considered, employers need a record of what the reviewer actually weighed. A reviewer who cannot say why they agreed with a system’s output will struggle to provide that explanation.

The hiring funnel is where the human line will be drawn

The carve-out for searching and shortlisting is significant. Automated screening can still drop candidates before a person is involved, so the protection starts at the final decision, not at the first filter. Employers will need to show where screening ends and the decision begins, and the explanation right means that boundary may have to be described to the worker. That boundary matters most for firms that buy recruitment software rather than build it. The decree treats a rejection at the screening stage differently from a decision not to hire at the end of the process, so vendors’ products may need to show which outputs feed a shortlist and which feed a final offer. Performance reviews and disciplinary steps have no such carve-out, so any tool that scores staff during employment falls under the full rule. For an overview of related rules, see TSN’s AI Regulation hub.

From 23 October, any Italian employer whose software produces a dismissal recommendation needs a named person ready to explain why they agreed with it.

Sources

• Gazzetta Ufficiale: Legislative Decree 7 October 2026, no. 179, record and entry into force (in Italian)
• Gazzetta Ufficiale Serie Generale no. 234, 8 October 2026, full issue PDF (in Italian)
• Normattiva: Law 23 September 2025, no. 132, Article 11 (in Italian)
• PwC Italia: Artificial intelligence in the workplace: what changes from 23 October 2026

Related stories

• AI Regulation: EU AI Act, UK Rules and US Moves | TSN Hub
• EU AI Act Gets Its First Harmonised Standard, for Quality Management Systems
• South Korean Bill Would Make Firms Report AI-Driven Mass Layoffs, JoongAng Says