Italy has become the first EU member state to pass a national artificial intelligence law which secured final parliamentary approval on 17 September 2025 and complements the EU AI Act. This domestic legislation follows on from Italy's "National AI Strategy" which, already back in 2020, highlighted that an ethical regulatory framework for AI must ensure transparency, accountability, and reliability in order to stimulate citizens' trust and engagement in a thriving AI ecosystem.
The law aims to set human‑centric guardrails around AI deployment, while encouraging innovation across the economy. Its mix of sectoral rules, criminal law updates, copyright clarifications, and institutional choices offers a playbook (and some cautionary tales) for other countries, including the UK as it calibrates its own path.
Summary of the key provisions
- Human oversight and traceability: The law requires that AI‑assisted decisions remain subject to human oversight and traceability. For example, in healthcare, medical professionals (and not AI) must be the ones taking decisions relating to prevention, diagnosis, treatment, and therapeutic choices. Patients should also be made aware when medical professionals make use of AI technologies.
- Access by minors: Children under 14 may access AI only with parental consent.
- Criminal penalties: The law introduces a new offence for unlawful dissemination of AI‑generated or manipulated content (e.g., deepfakes), with prison terms of one to five years where unjust harm is caused, and increases penalties when AI is used to commit existing crimes, such as market manipulation.
- Governance and enforcement: Two existing governmental bodies - the Agency for Digital Italy (AgID) and the National Cybersecurity Agency (ACN) - are designated to enforce the law. Meanwhile, the Department for Digital Transformation will steer a national AI strategy.
- Investment signal: In addition, up to €1 billion is earmarked to support companies in AI, cybersecurity, quantum technologies, and telecoms.
- Copyright:
(1) Copyright protection of AI works. The law asserts that works created with "AI assistance" are eligible for protection if they result from "genuine human intellectual effort". This seems to be largely a clarification / re-statement of existing EU copyright principles which are generally understood to protect those elements of AI works that have resulted from true human creativity (e.g. where AI has been used as a tool to bring a human-led idea to life). Whether a work which has been wholly created by AI attracts copyright protection is still left open for legislative clarity at both EU and national level.
(2) Text-and-data mining (TDM). On the hotly-debated issue around TDM, the Italian law asserts that "text extractions or reproductions from works...on the Internet or in databases....for the purpose of extracting text and data through artificial intelligence models and systems, including generative intelligence" will be permitted, provided the principles of Italy's implementing legislation of the DSM Directive have been respected (e.g. that copyright owner "opt-outs" are respected and that the works were accessed "lawfully").
This would appear to be an attempt domestically to flow-through relevant copyright / TDM provisions from the EU AI Act which (as per Recital 105 and Article 53) essentially make clear that the Art 3 and 4 DSM copyright exceptions are relevant to the "development and training" of generative AI models.
However, the precise wording of the new Italian legislation is, on first reading, not limited to "AI training purposes" only. It would appear to permit TDM for wider AI use cases – which could include 'LLM memorisation' or 'retrieval augmented-generation' (RAG) – provided "opt outs" are respected and access to the work was "lawful". Perhaps inadvertently, therefore, the Italian law extends beyond the remits of what is envisaged as permissible under the EU AI Act.
It's worth also noting that Italy has implemented this new law against an uncertain backdrop where the EU Parliament are openly considering the possibility of tightening the TDM copyright exception under Art 4 of the DSM Directive to make it harder for AI companies to rely on this to train AI models. The tectonic plates of TDM legislation are still shifting and it remains to be seen how this new Italian law will square with EU-wide legislation, wherever it lands.
Why this matters for the UK
- Despite Italy's former Prime Minister, Mario Draghi's, call last week to pause the AI Act's high‑risk tier, Italy enacted a law to supplement those obligations, suggesting Member States are willing to move ahead nationally rather than wait for further EU‑level recalibration.
- Italy's law supplements EU AI Act principles with additional operational, sector‑level duties and granular criminal offences on top of the AI Act baseline. This adds to the EU's layered regulatory matrix comprising regulations, directives and national laws.
- Historically, UK regulators have largely advocated for a more flexible "pro-innovation approach", and the UK is likely to continue with its "wait and see" approach to AI regulation. Italy alone will not force a UK pivot but a cascade of similar national measures across the EU will raise pressure on the UK to align to reduce friction, particularly around criminal sanctions and protections for minors. However, the longer it takes for the UK to establish its stance on AI regulation, the more likely it is that it will become a follower of one of the emerging approaches.
- As readers will also no doubt remember, the UK's proposed extension to the TDM exception (which would have allowed widescale TDM by AI companies within the UK for AI training purposes) was shelved by the previous Government back in 2023 after the proposal received significant pushback from the creative industries. The current Government held a major consultation on copyright and AI last winter, with a preference expressed to introduce a new TDM exception to copyright in line with the provisions of Art 4 of the DSM Directive (essentially aligning the UK with EU law around TDM). While the consultation closed back in February, the Government is yet to issue a formal response and Italy's approach (for good or bad reasons) will certainly inform the Government's thinking when it comes to legislating on the issue.
- Meanwhile, companies offering AI into the EU may need to start planning for country‑by‑country obligations that may apply ahead of key AI Act deadlines, and adopt an "EU‑plus" baseline (AI Act core plus leading national add‑ons) to mitigate fragmentation and enforcement risk.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
