Italy Bans Fully Automated HR Decisions, Except Screening

From 23 October, Italy voids a dismissal that an algorithm decided on its own. The same rule does not reach the algorithm that screened a candidate out before anyone met them.

That gap sits in Article 41 of Legislative Decree 179/2026, which adapts Italian law to the AI Act. The article bans fully automated HR decisions on hiring, changing and ending employment, discipline and performance reviews included. It reserves the final word for a person with real power. Then it carves out candidate screening. For any employer that uses AI in HR in Italy, both halves matter, and the second half is the one that will surprise people.

What Article 41 requires

The decree is dated 7 October 2026 and appeared in the Gazzetta Ufficiale, no. 234, on 8 October. It enters into force on 23 October 2026. Article 41 applies when an employer uses an AI system, as the AI Act defines one, in decisions about the employment relationship.

In those cases, the employer must make sure that final decisions on hiring, changing or ending employment are not taken solely on the basis of automated processing. In short, fully automated HR decisions are out. The article names disciplinary measures and performance evaluations explicitly. The final decision is reserved to a natural person who exercises effective and autonomous power, in the decree’s own words “un potere effettivo e autonomo”.

Two further points close the obvious escape routes. Worker consent is no exception to the ban. And the rule applies without prejudice to Article 22 of the GDPR, so the existing right not to be subject to solely automated decisions stays in place alongside it.

The screening exception in automated HR decisions

The third sentence of Article 41(1) is the one to read twice. Research and selection of candidates do not count as final decisions on establishing the employment relationship. That holds even where screening means a candidate does not reach the next stage.

In practice, an AI tool that ranks a thousand applications and discards nine hundred falls outside this Italian rule on automated HR decisions. Automated HR decisions at the point of hiring are covered; automated HR decisions at the point of shortlisting are not. A person has to make the final hiring decision. Nobody has to have seen the candidates who never reached that person.

The carve-out is not a free pass, however. Article 22 of the GDPR can still apply to screening, because the decree leaves it untouched. What changes is the sanction. Under Article 41, the consequence of getting it wrong is that the decision is void. Under the GDPR, it is a complaint, an investigation and possibly a fine.

When automated HR decisions are void

Article 41(4) is short and severe. Any dismissal adopted in breach of the first paragraph is void, and so is any other decision adopted in breach of it.

That makes automated HR decisions a different problem from a GDPR fine. The employer pays a fine once. A void decision has to be unwound, along with everything built on it. And a performance review that fed a promotion round raises the same question about the round.

The decree also gives workers a way to test the claim. Before processing starts, the employer must meet its information duties under Italian employment law. On request, a worker can obtain, through a natural person, an intelligible explanation of a decision. That explanation includes whether and how far AI influenced it and the main parameters considered.

What effective and autonomous means

The decree does not define the phrase, and no Italian court has yet tested it. The plain reading is demanding. A manager who approves two hundred AI-ranked appraisals in an afternoon exercises a signature, not a power. Effective means the person can overturn the output; autonomous means nobody penalises them for doing so.

Our earlier piece on meaningful human involvement set out a three-part test drawn from the Uber deactivation fine: the reviewer can overturn the output, sees the reasons and has the time and standing to disagree. Italy has now written a version of that test into employment law, with nullity attached.

One EU law, national rules for HR

In April we argued that there is no single EU-wide rollout of an algorithmic management tool, only a separate rollout for each member state. Article 41 is that argument made concrete. An HR platform that runs automated HR decisions identically across Europe will now need an Italian setting.

The Commission’s own figures show why this matters. In its Flash Eurobarometer of 14 October, 90% of SMEs and mid-caps said people make the final decisions, and monitoring performance and productivity was among the top uses of AI, at 36%. Italy has turned that self-assessment on automated HR decisions into a legal test with a consequence.

Before the rule bites, an employer with staff in Italy has four questions to answer:

• Which HR decisions in Italy draw on AI output today, such as appraisals, scheduling or dismissal files?
• For each one, who makes the final decision, and can that person overturn the output without consequence?
• Did workers receive the required information before processing started?
• Who can give a worker an intelligible explanation on request, including how far AI shaped the decision?

Employment lawyers will be asked what effective and autonomous means long before a court rules on it. The IAPP is launching an AI Specialist, Legal certification for lawyers, paralegals and legal operations professionals, with certification opening in January 2027. Our sister platform 22Academy already has a course for it.

For everyone else, the Italian rule on automated HR decisions leaves one question that no vendor can answer for you. When the system recommends, who in your organisation actually decides?