Italy’s Law 132/2025: A National AI Statute Inside the EU AI Act

Last verified against primary sources: 25 September 2026. Most coverage of European AI regulation stops at Regulation (EU) 2024/1689. That is a reasonable instinct — it is a regulation, not a directive, so it applies directly in every member state without national transposition. But it is not the whole European picture, and Italy is the cleanest available demonstration of why. On 23 September 2025 the Italian Parliament passed Legge 23 settembre 2025, n. 132, “Disposizioni e deleghe al Governo in materia di intelligenza artificiale.” It was published in Gazzetta Ufficiale Serie Generale n. 223 on 25 September 2025 (codice redazionale 25G00143) and entered into force on 10 October 2025.

What makes the law worth reading is not that Italy legislated on AI. It is what a member state legislated about, given that the Regulation already occupies the field of harmonised product rules. Law 132/2025 is careful not to compete with Brussels. Article 1(2) states the point outright: the law’s provisions “si interpretano e si applicano conformemente al regolamento (UE) 2024/1689” — they are to be interpreted and applied in conformity with the Regulation. Article 2 does not even write its own definition of an AI system; it adopts the one in Article 3(1) of the Regulation by reference. The statute then goes and does the things the Regulation does not do.

This page walks the structure. It is written from the consolidated Italian text on Normattiva and the Gazzetta Ufficiale original, not from secondary summaries.

The Shape of the Law: Six Chapters, 28 Articles

The statute is short by Italian standards — 28 articles across six chapters. The chapter headings alone tell you the division of labour:

Read as a whole, four things are being done here that the AI Act deliberately leaves to member states: sectoral rules for domestic institutions, designation of national authorities, copyright and criminal law, and national industrial and research strategy. None of those is a harmonised product requirement. All of them are squarely national competence.

Chapter II: Sectoral Provisions, Where the Law Is Most Concrete

Chapter II is the longest chapter and the one most likely to affect an organisation actually operating in Italy.

Health (arts. 7–10). Article 7 sets the frame: AI in healthcare may not select or condition access to care on discriminatory criteria; the patient has a right to be informed that AI technologies are in use; and the clinical decision remains “sempre rimessa agli esercenti la professione medica” — always reserved to the medical professional. Systems and the data behind them must be reliable, periodically verified and updated.

Work (arts. 11–12). Article 11 requires the employer or principal to inform the worker about AI use, hooking into the existing information obligations at article 1-bis of legislative decree 152/1997 on automated decision-making and monitoring systems. Article 12 creates an Osservatorio on AI adoption in the labour market at the Ministry of Labour, with its composition and functions to be set by ministerial decree within ninety days of entry into force.

Intellectual professions (art. 13). Two sentences that are easy to skim past and rather consequential: AI use in the intellectual professions is “finalizzato al solo esercizio delle attività strumentali e di supporto” — limited to instrumental and supporting activities, with the intellectual work itself predominating — and information about the AI systems the professional uses must be communicated to the client in clear, simple and exhaustive language.

Public administration (art. 14). Public bodies may use AI to speed up procedures, but only in a supporting role: the human being “resta l’unica responsabile dei provvedimenti e dei procedimenti” — remains solely responsible for the measures and procedures in which AI was used — and the administration must guarantee knowability of how the system works and traceability of its use. Article 14(4) adds the usual Italian sting: do it with existing resources.

Justice (art. 15). Every decision on the interpretation and application of law, on the assessment of facts and evidence, and on the adoption of measures is reserved to the magistrate. Until the Regulation is fully implemented, experimentation with and use of AI systems in ordinary judicial offices must be authorised by the Ministry of Justice, having heard the national authorities designated under article 20.

Civil procedure (art. 17). A single-line amendment with real litigation consequences: article 9 of the code of civil procedure is amended so that the tribunale has exclusive competence for cases “che hanno ad oggetto il funzionamento di un sistema di intelligenza artificiale” — cases concerning the functioning of an AI system. Italy has quietly created a jurisdictional rule for AI disputes.

Chapter III: Who Enforces What

Article 20 is the provision most people outside Italy will need. It designates AgID (Agenzia per l’Italia digitale) and ACN (Agenzia per la cybersicurezza nazionale) as the national AI authorities, and splits their roles:

• AgID promotes innovation and development, and runs notification, assessment, accreditation and monitoring of the bodies that verify AI system conformity. Article 20(2) designates AgID as the notifying authority under Article 70 of the Regulation.
• ACN is responsible for supervision — expressly including inspection and sanctioning — and is designated as the market surveillance authority and single point of contact with the EU institutions under the same Article 70.
• Banca d’Italia, CONSOB and IVASS keep the market surveillance role assigned to financial-sector supervisors by Article 74(6) of the Regulation. That carve-out is repeated twice in article 20, which suggests somebody fought for it.
• AgID and ACN must jointly establish regulatory sandboxes, consulting the Ministry of Defence on dual-use systems and the Ministry of Justice on judicial applications.
• Article 20(4) preserves the powers of the Garante per la protezione dei dati personali and of AGCOM as Digital Services Coordinator. Italy did not fold data protection into the AI regulator.

Article 19 puts the national AI strategy with the Presidency of the Council’s digital-transition structure, to be approved at least every two years by the interministerial committee for digital transition (CITD), with the Ministry of University and Research consulted on higher education and research and the Ministry of Defence on dual-use systems. Monitoring results go to Parliament annually. Article 19 has already been amended twice since entry into force — Normattiva records versions in force from 31 October 2025 and 31 December 2025, one of which added the Minister of Labour and Social Policies to the coordination committee created by article 19(6).

Article 23 authorises up to €1 billion of equity and quasi-equity investment in Italian AI, cybersecurity, quantum and telecoms companies through the existing venture-capital support fund. Article 21 is much smaller and oddly specific: €300,000 per year for 2025 and 2026 for AI pilots in foreign-ministry consular services. Article 27 makes the rest of the law cost-neutral by decree.

Chapter IV: Copyright, and the Word “Umano”

Article 25 does two things to Italy’s 1941 copyright law, and both are short enough to quote.

First, it inserts the word umano into article 1: protection now attaches to “opere dell’ingegno umano… anche laddove create con l’ausilio di strumenti di intelligenza artificiale, purché costituenti risultato del lavoro intellettuale dell’autore” — works of human intellect, including where created with the aid of AI tools, provided they constitute the result of the author’s intellectual work. That is a legislated answer to a question most jurisdictions are still resolving through litigation and registry practice: AI assistance does not defeat protection, but the human intellectual contribution has to be there.

Second, it adds a new article 70-septies permitting reproduction and extraction from works and other material lawfully accessed online or in databases, for text and data mining through AI systems including generative ones, in conformity with articles 70-ter and 70-quater — the existing Italian TDM exceptions, the first of which is the research-organisation exception. The provision does not create a new mining right. It confirms that the existing exceptions cover AI training, subject to their existing conditions.

Chapter V: The Criminal Provisions

Article 26 is where Law 132/2025 stops being administrative. It makes five changes:

• A new general aggravating circumstance at article 61 no. 11-decies of the criminal code: committing an offence through the use of AI systems, where those systems constituted an insidious means, obstructed public or private defence, or aggravated the consequences of the offence.
• A new offence at article 612-quater, illecita diffusione di contenuti generati o alterati con sistemi di intelligenza artificiale: causing unjust harm by transferring, publishing or otherwise disseminating, without consent, AI-falsified or AI-altered images, video or voice capable of deceiving as to their genuineness. Punishable by one to five years imprisonment, on complaint by the injured person — but prosecuted d’ufficio where connected to another such offence, or where committed against a person incapable by age or infirmity, or against a public authority because of its functions.
• Article 294 (offences against citizens’ political rights) gains a two-to-six-year penalty where the deceit is carried out using AI systems.
• Market manipulation is aggravated: article 2637 of the civil code rises to two to seven years where committed through AI systems, as does article 185 of the consolidated financial law, with a fine of €25,000 to €6 million.
• Copyright enforcement: a new article 171(1)(a-ter) of law 633/1941 criminalises reproducing or extracting text or data in violation of articles 70-ter and 70-quater, expressly including through AI systems — the enforcement counterpart to article 25.

These are national criminal provisions. Nothing in the AI Act does this, and nothing in the AI Act stops a member state from doing it.

The Honest Caveat: Much of This Is a Delegation, Not a Rule

The law’s own title says it: “Disposizioni e deleghe al Governo” — provisions and delegations to the Government. A significant share of the substantive content is not in the statute at all. It is in decreti legislativi that the Government is authorised to adopt later, within statutory deadlines and against stated principi e criteri direttivi. Anyone reading Law 132/2025 as a finished rulebook will misread it.

The main delegations:

Article 16 is the most structurally interesting of these, because of a clause that is easy to miss. The delegated regime for training data must be built “senza obblighi ulteriori, negli ambiti soggetti al regolamento (UE) 2024/1689, rispetto a quanto già ivi stabilito” — without obligations additional to those the Regulation already establishes, in the areas the Regulation covers. That is an explicit anti-gold-plating instruction written into the delegation itself. Article 16 also directs that remedies and a sanctions regime be created, and that disputes go to the specialised business sections of the courts.

Article 24’s criteria are worth reading in full if you work in Italy, because they are where the Government is told what to build: full supervisory, inspection and sanctioning powers for the article 20 authorities; penalties within the limits of Article 99 of the Regulation; AI literacy and training programmes; training by professional orders, with a possible equo compenso modulated by AI-related responsibility and risk; STEM curriculum strengthening; a dedicated regime for AI in policing; and remote or on-site inspection powers, including unannounced, over high-risk systems and real-world testing.

The law is also still moving. Article 24 gained a new paragraph 5-bis, in force from 27 June 2026, authorising up to €100 million from the national schools programme (PN scuola 2021–2027) for the AI-literacy criterion — a rare crack in the statute’s financial-invariance discipline. Article 26 was itself renumbered by a later statute.

As at 25 September 2026, the twelve-month delegation deadline is roughly two weeks away. We have verified the deadline from the consolidated text; we have not verified from primary sources whether each delegated decree has been adopted, and readers who need that answer should check the Gazzetta Ufficiale directly rather than rely on this page.

Why This Matters If You Do Not Operate in Italy

Three transferable lessons, none of which depends on caring about Italian law specifically.

One: “the EU AI Act” is not the complete European compliance surface. A directly applicable regulation harmonises product requirements. It does not harmonise criminal law, copyright, professional regulation, civil jurisdiction, public-administration procedure, or health-data governance. Law 132/2025 legislates in all six. An organisation that has mapped its obligations to the Regulation and stopped has mapped one layer of two. The same will be true, with different content, in every other member state that legislates.

Two: the designation layer is where enforcement actually lives. The Regulation requires member states to designate notifying and market surveillance authorities; it does not name them. Until a member state does, there is nobody to inspect you. Article 20 is the provision that turns the Regulation from text into a body with inspectors. For a comparative view of how differently jurisdictions build that layer, see our comparison of frontier AI law across ten jurisdictions and the broader jurisdiction map of AI regulations around the world.

Three: an enabling statute is a forecast, not a rule. The most useful thing to extract from Law 132/2025 today is its delegation calendar, not its prohibitions. The binding detail on training data, on sanctions, and on unlawful AI use will arrive in decrees. This is a structural feature of Italian legislative technique, and an argument for reading any national AI law’s deleghe before its principles. Compare the supranational instrument covered in our guide to the Council of Europe AI Treaty (CETS 225), which has the opposite problem: obligations that exist on paper and depend entirely on how each ratifying state implements them.

Research-Administration Relevance

The research angle here is narrow but genuine, and it runs through three provisions.

Articles 8 and 9: health-AI research data. This is the most consequential part of the law for a research office. Article 8 declares that processing of data — including personal data — by public bodies, non-profit private bodies, IRCCS (Italy’s scientific hospitalisation and care institutes) and private health-sector bodies participating in projects with them, for research and scientific experimentation in building AI systems for prevention, diagnosis, treatment, drug development, rehabilitation technology, medical devices and public health, is “di rilevante interesse pubblico” — of substantial public interest — under Article 9(2)(g) GDPR. Article 8(2) then authorises secondary use of personal data stripped of direct identifiers, without further consent, subject to an information notice that may be given generally on the controller’s website. Article 8(3) permits processing for anonymisation, pseudonymisation and synthesis. AGENAS may issue anonymisation and synthetic-data guidelines after an opinion from the Garante. Article 8(5) requires these processing operations to be notified to the Garante with the full Articles 24, 25, 32 and 35 GDPR information. Article 9 then hands the Ministry of Health a 120-day decree to set up simplified processing and research sandboxes.

For a sponsored-programmes or human-subjects office, that is a real change in the legal basis and consent posture for AI-related health research in Italy — a route that does not run through fresh consent, but does run through a notification duty and an as-yet-unsettled ministerial decree. Anyone relying on it should read articles 8 and 9 in the original.

Article 24(2)(l): universities, research bodies and technology transfer. One of the express delegation criteria is the valorizzazione of AI research and technology-transfer activity carried out by universities, AFAM institutions, ITS Academy technical institutes and public research bodies — specifically, by making it easier to involve the university and research system in creating and using regulatory sandboxes together with industry, and by simplifying collaboration instruments between that system and the article 20 authorities. Whatever the decrees eventually say, the statute has put research-system participation in AI sandboxes on the Government’s required agenda.

Articles 13, 14 and 25. Italian public universities and public research bodies are public administrations, so article 14’s knowability, traceability and sole-human-responsibility requirements apply to their administrative use of AI. Article 13’s professional-disclosure duty reaches the regulated professions that research institutions employ and contract with. And article 25’s confirmation that the existing TDM exceptions — including the research-organisation exception at article 70-ter — cover AI training is directly relevant to any Italian research organisation mining licensed or openly accessible corpora, particularly now that article 26(3) attaches criminal liability to mining that falls outside those exceptions.

What This Page Does Not Claim

Law 132/2025 is frequently described as the first national AI statute adopted by an EU member state. We have not verified that characterisation against a primary source and do not assert it here. What is verifiable from the Gazzetta Ufficiale and Normattiva is narrower and sufficient: Italy adopted a comprehensive national AI statute that took effect on 10 October 2025, expressly subordinate to Regulation (EU) 2024/1689, and it did so by legislating around the Regulation rather than on top of it.

Frequently Asked Questions

Does Italy’s Law 132/2025 replace or override the EU AI Act?

No. Article 1(2) states that the law’s provisions are interpreted and applied in conformity with Regulation (EU) 2024/1689, and article 2 adopts the Regulation’s own definition of an AI system by reference. Article 16 goes further, instructing that the delegated regime on training data impose no obligations additional to those the Regulation already sets in the areas it covers. The statute is designed to sit inside the harmonised regime, not alongside or above it.

Who is Italy’s AI regulator?

There are two, plus three financial supervisors. Article 20 designates AgID and ACN as the national AI authorities: AgID as the notifying authority under Article 70 of the Regulation and the body handling conformity-assessment accreditation; ACN as the market surveillance authority, single EU point of contact, and the agency with inspection and sanctioning powers. Banca d’Italia, CONSOB and IVASS retain market surveillance for the financial sector under Article 74(6). The Garante and AGCOM keep their existing powers unchanged.

When do the implementing decrees have to be adopted?

The three legislative-decree delegations — article 16 on training data, article 24(1) on aligning national law to the Regulation, and article 24(3) on unlawful AI use — each run twelve months from entry into force, expiring on 10 October 2026, with a sixty-day extension mechanism where parliamentary opinions fall late. Two ministerial decrees had shorter deadlines: ninety days for the labour-market Observatory (article 12) and one hundred and twenty days for health-research data processing (article 9).

What criminal penalties does the law create for deepfakes?

Article 26 inserts a new article 612-quater into the criminal code punishing unlawful dissemination of AI-generated or AI-altered images, video or voice that cause unjust harm, with one to five years’ imprisonment. It is normally prosecuted on the victim’s complaint, but prosecuted automatically where it is connected to another automatically prosecuted offence or committed against a person incapable by age or infirmity or against a public authority because of its functions. Separately, using AI systems to commit any offence can now be a general aggravating circumstance under article 61 no. 11-decies.

Does the law change copyright in AI-generated works?

Yes, in a limited and specific way. Article 25 amends article 1 of law 633/1941 so that protection covers works of human intellect, including works created with the aid of AI tools, provided they are the result of the author’s intellectual work. It also adds article 70-septies confirming that text and data mining through AI systems from lawfully accessed material is permitted in conformity with the existing exceptions at articles 70-ter and 70-quater — and article 26(3) makes mining outside those exceptions a criminal offence.

Does it apply to organisations without an Italian establishment?

The statute does not create a general extraterritorial compliance regime in the way the Regulation does. Its sectoral provisions bind Italian public administrations, Italian healthcare and research bodies, employers and professionals operating in Italy, and its criminal provisions operate under ordinary Italian criminal jurisdiction. The parts that reach a foreign provider are the enforcement machinery in articles 20 and 24 — because those are the authorities that will exercise the Regulation’s own supervisory and sanctioning powers in Italy.

Primary Sources

• Legge 23 settembre 2025, n. 132, Disposizioni e deleghe al Governo in materia di intelligenza artificiale, Gazzetta Ufficiale Serie Generale n. 223, 25 September 2025 (25G00143), in force 10 October 2025 — Gazzetta Ufficiale, original text.
• The consolidated multi-version text, including the amendments to articles 19, 24 and 26 — Normattiva, Legge 132/2025.

Related reading

• Germany’s KI-MIG: The National AI Act Implementation Statute
• Italy’s D.Lgs. 160/2026: Police AI, Biometrics and AI Liability
• Vietnam’s Law on Artificial Intelligence No. 134/2025/QH15
• EU AI Act Article 70: France, Germany, Spain, and Ireland Compared
• China’s AI Regulation: Interim Measures to the 2025 Labeling Rules
• NYC Local Law 144 vs. California’s FEHA Automated-Decision System Rules