Italy’s Decree 179/2026 Completes Its AI Act Framework: Regulators, Fines and the National Sandbox

Italy finished the second half of its national AI Act framework on October 23, 2026, when Legislative Decree No. 179 of October 7, 2026 entered into force. The decree was published in the Official Journal (Gazzetta Ufficiale) No. 234 on October 8. It follows Law No. 132 of September 23, 2025, Italy’s general AI law, which named the national AI authorities and asked the government to align Italian law with the EU AI Act. The new decree fills in the parts the EU regulation leaves to member states: who supervises, who fines, how much, and how the national regulatory sandbox will work.

The EU AI Act applies directly across the bloc, so the decree does not rewrite the obligations on providers and deployers. It changes the practical picture for anyone selling or using AI systems in Italy, because it decides which regulator knocks on the door and which court hears the appeal. In two places Italy has also gone further than the EU text.

Who does what under Decree 179

The decree has 54 articles in five chapters. The longest chapter, Articles 2 to 35, deals with authorities, enforcement and sandboxes. Later chapters cover AI in training and education, the professions, work, health care and public administration, plus an amendment to Italy’s trade secrets provision.

• Agency for Digital Italy (AgID) is the notifying authority. It designates and monitors the conformity assessment bodies that check high-risk systems, and it can restrict, suspend or withdraw a designation.
• National Cybersecurity Agency (ACN) is the general market surveillance authority for AI systems and Italy’s single point of contact with EU institutions. It has inspection powers, including over real-world testing, and can act without prior notice.
• Bank of Italy, Consob and IVASS supervise high-risk AI used by the banks, investment firms and insurers they already regulate.
• The Garante, Italy’s data protection authority, has a narrower role tied to Article 74(8) of the AI Act, which covers certain high-risk uses such as law enforcement and border management. Late changes to the decree gave it more room, including the power to issue guidelines in that area.

Articles 7 to 10 are devoted to cooperation among these bodies, and a coordination committee brings them together. Italian commentators, including an analysis published by Altalex on October 13, have pointed out that this cooperation will take real work and cannot be assumed.

Penalties, and where Italy went further than the EU

Article 23 sets four tiers of fines that track Article 99 of the AI Act. Our overview of EU AI Act penalties covers the EU baseline. The Italian tiers are:

1. Up to €35 million or 7% of worldwide annual turnover, whichever is higher, for prohibited AI practices under Article 5 of the AI Act.
2. Up to €15 million or 3% for breaches of obligations on providers, authorized representatives, importers, distributors, deployers and notified bodies, and for transparency breaches. Italy added two items here that Article 99 does not list: failures in serious incident reporting (Article 73) and in the procedure for systems a provider has classified as not high-risk (Article 80).
3. Up to €7.5 million or 1% for supplying incorrect, incomplete or misleading information to notified bodies or authorities.
4. Up to €1 million or 0.5% for breaching the duty to carry out a fundamental rights impact assessment under Article 27 or the right to an explanation of individual decisions under Article 86. This tier is a national choice. Article 99 does not list these breaches.

Small and medium-sized enterprises, micro-enterprises and innovative start-ups get the AI Act’s reduction, under which the cap is the lower of the fixed amount and the percentage of turnover. An earlier draft halved fines for individuals, but that was dropped, so a sole trader faces the same range as a company unless the SME rule applies. For minor breaches, Article 22 lets an authority order the breach fixed by a deadline, or publish a public statement naming the violation, instead of fining. Ignoring the order can raise the eventual fine by up to a third.

Timing matters. Many AI Act obligations do not apply yet, and the EU’s later amendments moved some high-risk deadlines. Our AI Act deadlines page and our report on the EU’s last-minute deal to rewrite parts of the AI Act explain the shifting dates. The Altalex commentary notes that for now the only Italian penalties that can actually bite are those for prohibited practices, misleading information, and the Article 50 transparency duties. Like the EU text as amended, the decree sets no fine for breaching the AI literacy duty in Article 4, although literacy failures can count when a fine is calculated.

One quirk deserves attention. Appeals go to different courts depending on who issued the fine: the administrative courts for ACN, the ordinary courts for AgID and the Garante, and the courts of appeal for the Bank of Italy and Consob. The same breach could be judged in different forums. Fine revenue is also returned to the authorities’ own budgets, so enforcement partly funds itself.

The national AI sandbox

Articles 26 to 31 set up the Italian AI sandbox, the controlled testing space that Article 57 of the AI Act requires each member state to run. AgID and ACN run it jointly, with priority for SMEs and start-ups. The decree creates the sandbox as of October 23, 2026, but access rules, operations, eligibility criteria and the make-up of the operating group are left to joint decrees from the two agencies’ directors, and the decree sets no deadline for them. The Garante can join operating group meetings for projects that process personal data. Separate sandboxes for dual-use systems and for AI used in the courts can be created by prime ministerial decree, and regions can set up their own.

What organizations operating in Italy should do now

• Map your systems to the right regulator. A bank’s credit-scoring model falls to the Bank of Italy. A general-purpose chatbot sold to Italian businesses falls to ACN. Knowing this tells you whose guidance to watch. Our guide to AI Act risk categories is a starting point for classification.
• Treat incident reporting as a fineable duty. Because Italy put Article 73 failures in the €15 million tier, providers of high-risk systems should have a written process for spotting, escalating and reporting serious incidents.
• Document fundamental rights assessments. Deployers covered by Article 27 face a national fine if they skip the assessment. A completed assessment is evidence that you asked the questions, not proof that the system is safe.
• Plan for explanation requests. Article 86 gives people affected by certain high-risk decisions a right to an explanation, and Italy attached a fine to it. Decide who answers these requests and what records they need.
• Watch for the implementing decrees. Inspection procedures, complaint handling and sandbox access all depend on decrees not yet adopted.

Non-EU companies should not assume distance protects them. As our piece on AI Act compliance for U.S. companies explains, placing a system on the EU market or using its output in the EU can bring a company into scope.

How Captain Compliance can help

Captain Compliance offers AI governance advisory work, including governance structures, transparency documentation and third-party AI assessments, along with assessment tools that help teams record what an AI system does, what data it uses and who is accountable. That record is useful when an Italian regulator asks questions. It does not guarantee compliance with the AI Act or the decree.

Frequently asked questions

When did Legislative Decree 179/2026 take effect?

It entered into force on October 23, 2026, after publication in the Official Journal on October 8, 2026.

Who is Italy’s main AI regulator?

The National Cybersecurity Agency (ACN) is the general market surveillance authority and single point of contact. AgID is the notifying authority. Financial regulators cover high-risk AI in their sectors, and the Garante has a narrower role.

Are Italy’s AI fines higher than the EU’s?

The maximum amounts match the AI Act. Italy added breaches to the tiers: serious incident reporting and the non-high-risk classification procedure in the 3% tier, and fundamental rights assessments and the right to explanation in a new 0.5% tier.

Is the Italian AI sandbox open now?

It legally exists from October 23, 2026, but access depends on joint decrees from AgID and ACN that have not yet been adopted.

Does the decree change GDPR obligations?

No. GDPR still applies to personal data used in AI systems, and the decree preserves the Garante’s powers.

The paperwork is now the risk

Italy’s choices point in one direction. Several of the duties it chose to back with national fines are about records: incident reports, classification decisions, rights assessments and explanations. Those are the things a regulator can ask to see on day one of an inquiry. A company that cannot produce them may find that the first Italian AI fine is not about what its model did, but about what it failed to write down.