Italy has a general national AI law that runs alongside the EU AI Act, and since autumn 2026 it has teeth: new crimes, a causation presumption for AI harm and a national AI supervisor that is also the cyber agency. Law no. 132 of 23 September 2025 sets national principles, sector rules, authorities and criminal offences, and delegates the Government to adapt Italian law to Regulation (EU) 2024/1689. Two legislative decrees made under that delegation, Legislative Decree 160/2026 and Legislative Decree 179/2026, are now published and in force. This guide explains what the law and decrees require, who supervises, the penalties, and what security, AI governance and identity teams operating in Italy need to do. Status line: checked against the Gazzetta Ufficiale, Normattiva and legal commentary on 11 November 2026.
Source note: provisions come from Law 132/2025 in the Gazzetta Ufficiale and Normattiva; details of Legislative Decrees 160/2026 and 179/2026 not taken from the official text are marked as reported. Status is as of November 2026. Confirm against the official text before relying on it.
Key takeaways
- Law 132/2025 sits beside the AI Act, not above it. Article 1 requires the law to be interpreted and applied in line with Regulation (EU) 2024/1689, and Article 3(5) states it creates no obligations beyond that Regulation for AI systems and general-purpose models.
- Cybersecurity is a precondition, in law. Article 3(6) makes cybersecurity across the whole lifecycle of AI systems and models an essential precondition, including resilience against attempts to alter their use, behaviour, performance or safety settings.
- ACN supervises AI. Article 20 designates the National Cybersecurity Agency (ACN) as the market surveillance authority with inspection and sanction powers and single point of contact, and AgID as the notifying authority; Banca d'Italia, CONSOB and IVASS keep financial sector supervision.
- Criminal exposure is real. Article 26 created the deepfake offence in Article 612-quater of the Criminal Code (one to five years), and Legislative Decree 160/2026 added Article 437-bis for omitting safety measures on, or unlawfully altering, high-risk AI systems.
- Civil claims got easier. Legislative Decree 160/2026 lets courts order disclosure of how an AI system works and presumes causation where harm follows a breach of AI Act obligations, and AI Act compliance alone does not exclude liability.
Scope: who and what is covered
Law 132/2025 was published in Gazzetta Ufficiale Serie Generale no. 223 on 25 September 2025 and entered into force on 10 October 2025. Article 1 says it sets principles for the research, testing, development, adoption and use of AI systems and models, promoting correct, transparent and responsible use in a human-centred approach. It is organised in six chapters: principles (Articles 1 to 6), sector provisions (7 to 18), national strategy and authorities (19 to 24), user protection and copyright (25), criminal provisions (26) and final provisions (27 to 28).
- Who: anyone developing, providing, deploying or using AI in Italy, including employers, professionals, public administrations, the judiciary and healthcare bodies. Most private-sector obligations flow through the AI Act categories of provider and deployer.
- Exclusions: Article 6 excludes activities for national security, defence and certain intelligence and police functions, while still requiring respect for constitutional rights.
- Minors: under Article 4, children under 14 need consent from whoever holds parental responsibility to access AI technologies and for related processing of personal data; minors aged 14 to 17 may consent themselves if the information is easily accessible and understandable.
| Instrument | Date and publication | In force | Main content |
|---|---|---|---|
| Law 132/2025 | 23 September 2025, GU no. 223 of 25 September 2025 | 10 October 2025 | Principles, sector rules, authorities, deepfake and AI aggravating offences, delegations |
| Legislative Decree 160/2026 | 9 September 2026, GU no. 214 of 15 September 2026 | 30 September 2026 | Police use of AI and biometrics, new Article 437-bis offence, corporate liability, civil liability rules |
| Legislative Decree 179/2026 | 7 October 2026, GU no. 234 of 8 October 2026 | 23 October 2026 (as reported) | Powers of national authorities, administrative sanctions, AI training, workplace risk assessment |
Key obligations
Sector duties in Law 132/2025
| Article | Who | Obligation (summarised from the Gazzetta Ufficiale text) |
|---|---|---|
| Art. 3 | All | Respect fundamental rights, human autonomy, transparency, explainability and human oversight; cybersecurity across the lifecycle as a precondition (para. 6) |
| Art. 4 | Services offered to minors | Parental consent under 14; understandable information for 14 to 17 |
| Art. 11 | Employers and clients | Inform workers when AI is used, under Article 1-bis of Legislative Decree 152/1997; no discrimination on listed grounds |
| Art. 13 | Intellectual professions | AI only in a supporting role with intellectual work predominant; explain the AI tools used to clients in clear, simple and complete terms |
| Art. 14 | Public administration | AI to improve efficiency with transparency and traceability; a human remains solely responsible for decisions |
| Art. 15 | Judiciary | Interpretation, assessment of facts and evidence and rulings reserved to magistrates |
| Art. 25 | Rights holders and model developers | Copyright protects human works made with AI tools if they result from the author's intellectual effort; new Article 70-septies on text and data mining with AI, within Articles 70-ter and 70-quater |
Security-relevant duties from the implementing decrees
- Safety and oversight measures on high-risk AI: Article 437-bis of the Criminal Code, inserted by Legislative Decree 160/2026, makes it an offence to omit technical safety or human-oversight measures on high-risk AI systems where this creates danger to life or to public or individual safety, and to alter high-risk AI systems unlawfully. This turns AI Act Article 14 and Article 15 controls into potential criminal exposure. See the EU AI Act Article 15 Guide.
- Workplace risk assessment: Sistema Penale reports that Article 42 of Legislative Decree 179/2026 requires AI systems affecting work organisation, pace, task performance or safety-relevant decisions to be assessed in the workplace risk assessment under Article 28 of Legislative Decree 81/2008.
- Corporate liability: Legislative Decree 160/2026 adds Articles 437-bis and 612-quater to Article 25-vicies of Legislative Decree 231/2001, so companies can be liable for these offences committed in their interest. Organisational models under Decree 231 need updating.
- Police biometrics: real-time remote biometric identification requires prior authorisation from the prosecutor, and identification of a targeted suspect by post-event facial recognition needs a preliminary investigations judge's authorisation.
Timeline and current status
| Date | Event |
|---|---|
| 23 September 2025 | Law 132/2025 signed |
| 25 September 2025 | Published in Gazzetta Ufficiale no. 223 |
| 10 October 2025 | Law 132/2025 enters into force |
| 10 June 2026 | Council of Ministers approves the first two draft implementing decrees (preliminary) |
| 26 June 2026 | Last update to the consolidated text shown on Normattiva (Articles 19 and 24 amended) |
| 4 August 2026 | Council of Ministers gives final approval to both decrees (session no. 185, per the press release as reported) |
| 30 September 2026 | Legislative Decree 160/2026 enters into force |
| 10 October 2026 | Deadline for the Article 24 delegation (12 months from entry into force) |
| 23 October 2026 | Legislative Decree 179/2026 enters into force (as reported) |
Article 24 also gave 12-month delegations on unlawful use of AI and related liability. Further decrees, guidance from ACN or AgID, and updates to the national AI strategy under Article 19 may follow. AI Act application dates themselves are set at EU level and have been adjusted; check the EU AI Act Implementation Timeline Tracker rather than relying on dates in Italian commentary.
Enforcement and penalties
| Provision | Conduct | Penalty |
|---|---|---|
| Art. 612-quater c.p. (Law 132/2025) | Unlawful dissemination of AI-generated or altered images, video or voice capable of deceiving | One to five years' imprisonment; prosecutable on complaint, with exceptions |
| Art. 61 no. 11-decies c.p. (Law 132/2025) | Using AI as an insidious means, hindering defence or worsening consequences of any offence | Aggravating circumstance |
| Art. 185 TUF (Law 132/2025) | Market manipulation committed through AI | Two to seven years and a fine of 25,000 to 6 million euro |
| Art. 437-bis c.p. (Decree 160/2026) | Omitting safety or human-oversight measures on high-risk AI creating danger | Imprisonment, with heavier penalties where State security is endangered and reduced penalties for negligence (as reported by Sistema Penale; confirm ranges in the decree text) |
| Art. 437-bis c.p. (Decree 160/2026) | Unlawfully altering a high-risk AI system creating danger | Imprisonment, heavier where State security is endangered (as reported by Sistema Penale; confirm ranges in the decree text) |
| Art. 23, Decree 179/2026 | Prohibited practices; operator obligations; misleading information to authorities | Administrative fines reported as aligned with the AI Act Article 99 tiers (Sistema Penale) |
| Art. 23, Decree 179/2026 | Breach of the fundamental rights impact assessment (AI Act Art. 27) or right to explanation (Art. 86) | Administrative fines set nationally (as reported by Sistema Penale; confirm amounts in the decree text) |
Administrative enforcement sits mainly with ACN, with AgID, the sector financial supervisors and the Garante for data protection in their areas. Civil claimants benefit from evidence disclosure (Decree 160/2026 Article 17), a presumption of causation where AI Act obligations were breached (Article 18), the rule that AI Act conformity does not by itself exclude liability (Article 19) and direct action against insurers (Article 20). See the AI Liability and Product Liability Guide.
What it means for AI agents and non-human identities
- Unlawful alteration covers credentials: Article 3(6) and Article 437-bis both target changes to an AI system's behaviour or safety settings. The practical attack path is a stolen API key, service account or pipeline token used to change prompts, tools, model weights or guardrail configuration. Protecting those identities is how you show the system was not unlawfully altered on your watch. See the Non-Human Identity (NHI) Guide.
- Human oversight must be provable: Article 437-bis penalises omitting human-oversight measures. For agents that act autonomously, keep approval gates for high-impact actions and logs that tie each action to an agent identity and the human who authorised its scope.
- Evidence will be ordered: courts can order disclosure of how a system works. Agent logs, tool-call records and configuration history are that evidence; if they are incomplete, the causation presumption works against you. See the Audit Readiness for AI Agents and NHIs.
- Deepfakes and identity fraud: Article 612-quater addresses deceptive synthetic media, which is also a route to help desk and payment fraud. See the Deepfake and AI Impersonation Guide.
Compliance steps
- Inventory (first month): list AI systems used or offered in Italy, mark AI Act role (provider, deployer, importer) and risk class, and flag any used on minors, workers, clients of professionals or in regulated finance.
- Update the Decree 231 model: add Articles 437-bis and 612-quater to the risk map and controls of your organisational model, with owners for high-risk AI safety measures.
- Secure the lifecycle: apply Article 3(6): integrity controls on models, prompts and configuration (signed artefacts, change control), and least-privilege, rotated credentials for every pipeline and agent identity. See the Italy NIS2 Implementation Guide where ACN also supervises you under NIS2.
- Prove oversight: document human-oversight measures for each high-risk system, test that they work, and log their use.
- Workplace and professional disclosures: issue worker notices under Article 11, include AI in the Decree 81/2008 risk assessment, and give professional clients plain-language AI disclosures under Article 13.
- Minors and personal data: implement age checks and parental consent flows for under-14s and align processing with the GDPR. See the Italy Data Protection Law guide.
- Prepare for litigation: retain technical documentation and logs in a form you can disclose, and review insurance given direct action against insurers.
- Test and roll back: pilot oversight and logging changes on one high-risk system, check evidence quality with legal, then extend. Keep the previous model and configuration versions so an unsafe change can be reverted with a record of who approved it.
Decision guidance
- Provider versus deployer: providers carry design obligations and the heaviest sanction tiers; deployers mainly carry use, oversight and FRIA duties. If you fine-tune or rebrand, check whether you have become a provider. See the EU AI Act Compliance Guide.
- Small organisations: focus on Article 11 notices, Article 13 disclosures, deepfake awareness and credential security for AI tools; most will not run high-risk systems.
- Large or regulated organisations: expect parallel supervision by ACN and Banca d'Italia, CONSOB or IVASS; align one evidence set to both.
Detection and measurement
- Coverage: percentage of AI systems in the inventory with an assigned AI Act role and risk class.
- Oversight evidence: share of high-risk systems with tested and logged human-oversight measures.
- Integrity: unauthorised configuration or model changes detected per quarter, and time to revert.
- Credential hygiene: percentage of AI pipeline and agent credentials with an owner and rotation within policy.
- Disclosure readiness: time to assemble a complete evidence pack for one system on request.
Framework mapping
| Italian requirement | NIST CSF 2.0 | NIST SP 800-53 Rev 5 |
|---|---|---|
| Know and track Law 132/2025 and decrees | GV.OC-03 | PM-28 Risk Framing |
| AI policy and accountability | GV.PO-01 | |
| Lifecycle cybersecurity, no unlawful alteration (Art. 3(6), Art. 437-bis) | SI-7 Software, Firmware, and Information Integrity | |
| Evidence for oversight and disclosure | AU-2 Event Logging | |
| Lawful processing for minors and workers | PT-2 Authority to Process Personally Identifiable Information |
Common failure patterns
- Treating the Italian law as a copy of the AI Act: the criminal offences, Decree 231 liability, worker notices and minors' consent are national additions.
- Oversight on paper only: a documented approval step that agents can bypass is the omission Article 437-bis targets.
- Shared AI platform keys: one key across teams makes unauthorised alteration impossible to attribute.
- Logs not kept or not retrievable: with a causation presumption in play, missing evidence hurts the defendant.
- Ignoring sector supervisors: financial entities assume ACN alone supervises and miss Banca d'Italia, CONSOB or IVASS expectations.
Practitioner checklist
- Record Law 132/2025, Decree 160/2026 and Decree 179/2026 in your legal register with in-force dates.
- Map every AI system used in Italy to an AI Act role and risk class.
- Add Articles 437-bis and 612-quater to the Decree 231 organisational model.
- Protect model, prompt and configuration integrity and every AI pipeline credential.
- Test and log human-oversight measures on high-risk systems.
- Issue worker notices and update the Decree 81/2008 workplace risk assessment.
- Implement parental consent for under-14 users of AI services.
- Keep disclosable technical documentation and logs, and review liability insurance.
- Monitor ACN and AgID for guidance and further Article 24 decrees.
Standards and references
- Gazzetta Ufficiale: Legge 23 settembre 2025, n. 132, Disposizioni e deleghe al Governo in materia di intelligenza artificiale
- Normattiva: Law 132/2025 consolidated text
- Normattiva: Decreto legislativo 9 settembre 2026, n. 160
- Sistema Penale: Published in the Gazzetta Ufficiale, the implementing decrees of Law 132/2025
- Cyber Security 360: AI Act, the Government accelerates, more powers to ACN
- Noze: Italian AI Act decrees, final approval
Related NHI Mgmt Group resources: EU AI Act Compliance Guide · Italy NIS2 Implementation Guide · AI Liability and Product Liability Guide · AI Kill Switch Act Guide. This guide summarises Italian Law 132/2025 and Legislative Decrees 160/2026 and 179/2026 as of 11 November 2026 and is not legal advice.