EU AI Act Compliance for Creative Industries in Italy: A Complete Guide for Film, Music, Art, and IP Professionals

Last updated: September 2026 — by Avv. Claudia Roggero, Rome Bar since 2003, entertainment and copyright law, DANDI.media. Co-author of “The human element: Italy rewrites Article 1 of its Copyright Act for the age of AI”, Entertainment Law Review (Sweet & Maxwell). Teaches AI and rights in cinema at the Civica Scuola di Cinema Luchino Visconti, Milan.

Italy is the first EU member state with comprehensive national AI legislation. Law 132/2025, in force since 10 October 2025, sits alongside the EU AI Act — and for anyone producing creative work in Italy it is the more consequential of the two, because it changed the Copyright Act itself.

Creative businesses now operate under three overlapping regimes at once: the AI Act’s transparency and risk obligations, Italy’s national implementation with its own criminal provisions, and a rewritten Article 1 of the Copyright Act that defines when AI-assisted work is protected at all.

The rewritten Article 1: when AI-assisted work is protected

This is the change that matters most, and the one least discussed outside Italy.

Law 132/2025 amended Article 1 of the Italian Copyright Act to clarify that works created with the assistance of artificial intelligence are protected where the human creative contribution is genuine. The legislature codified what had until then been a matter of interpretation: AI is a tool, and the question is whether a human made the creative choices.

Three practical consequences follow.

• Purely AI-generated output is not protected. Prompt engineering alone does not establish authorship. Content generated without meaningful human creative input falls outside copyright — which means anyone can use it, including your competitors.
• AI-assisted work is protected, with the human as author. Selection, arrangement, editing and creative direction can constitute the required contribution.
• The distinction has to be documented. Where protection is later contested, the question is evidential: what did the human actually decide? Productions that keep no record of the creative process have nothing to show.

We examined the reform and its implications for cross-border practice in “The human element: Italy rewrites Article 1 of its Copyright Act for the age of AI”, published in the Entertainment Law Review (Sweet & Maxwell).

Law 132/2025: what Italy added

Beyond the copyright amendment, the national law introduces obligations that do not appear in the AI Act.

Why this matters for a foreign producer. A US or UK production working in Italy is used to treating AI as a contractual and reputational question. Here it is also a criminal one, and the exposure attaches to dissemination — not to creation. That changes who in the chain carries the risk.

The AI Act framework and timeline

Regulation (EU) 2024/1689 entered into force on 1 August 2024 and applies in phases:

• 2 February 2025 — prohibited practices apply, including certain biometric categorisation and emotion recognition in workplaces and education;
• 2 August 2025 — obligations for general-purpose AI models take effect: technical documentation, copyright policy, and publication of a sufficiently detailed summary of training data;
• 2 August 2026 — the bulk of the remaining obligations, including high-risk system requirements.

The risk-based structure runs from prohibited, through high-risk, to limited-risk systems subject to transparency obligations, to minimal-risk systems with no specific requirements.

Most creative-industry AI sits in the limited-risk tier — content generation, image manipulation, voice synthesis — which means the operative obligation is transparency rather than conformity assessment. That is a lighter regime than many producers assume, but it is not nothing.

Transparency obligations under Article 50

Article 50 requires disclosure where users interact with an AI system or where content is artificially generated or manipulated. For creative businesses:

• AI-generated content must be marked in a machine-readable format detectable as artificially generated;
• Deepfakes must be disclosed — content resembling real persons, places or events;
• Chatbots and conversational systems must inform users they are interacting with AI;
• Emotion recognition and biometric categorisation require notification of the persons exposed.

The exception that matters editorially: where AI-generated content forms part of an evidently artistic, creative, satirical or fictional work, transparency obligations are limited to disclosure that does not hamper display or enjoyment of the work. It is a narrower carve-out than it appears — it governs how you disclose, not whether.

Text and data mining: the opt-out

Training AI on copyrighted material runs through the text and data mining exceptions transposing the DSM Directive, with a rightsholder opt-out for commercial uses.

Two sides to this, and creative businesses usually sit on both.

As a rightsholder: the opt-out has to be exercised, in a machine-readable form, to have effect. Catalogues that have taken no position are, in practice, available. This is a live decision, not a default.

As a user of AI tools: the licence you hold from a provider says nothing about whether the training was lawful. Where the model was trained on material subject to an unexercised opt-out, the position is one thing; where the opt-out was exercised, another. Training data verification is now part of tool procurement, not an academic question.

The GPAI obligation to publish a training data summary, in force since August 2025, is what makes this checkable at all.

The tax credit AI clause

Audiovisual productions accessing the Italian cinema tax credit face a further requirement, and which provision applies depends on the scheme:

Under Article 9(4), contracts with authors, performers and executants must contain, on pain of inadmissibility, clauses allowing them to withhold consent to AI exploitation of their work, image or performance. Two companion obligations apply: a declaration at final application of which parts of the work used AI, and the non-eligibility of AI costs replacing creative or artistic performance.

The problem in practice is sequencing. Talent agreements are signed months before anyone opens the tax credit file, and “on pain of inadmissibility” is not a curable defect. See our Italian film tax credit guide.

Sector application

Film and audiovisual

AI in VFX, de-aging, voice synthesis, dubbing and score generation. Obligations: transparency labelling, performer consent for likeness and voice, the mandatory tax credit clause, and — where the treatment modifies a performance — the inalienable moral rights of the four statutory co-authors under Article 44 of the Copyright Act. See our moral rights in film guide.

Music

AI composition, mastering, voice cloning and stem separation. The distinctive issue is that a voice is protected as an attribute of personality independently of copyright: a licence on the composition says nothing about the right to imitate a singer. Commercial release imitating a named artist requires that artist’s express consent.

Publishing

AI-assisted writing, translation and editorial production. The Article 1 question is central: where does human creative contribution lie, and can it be evidenced?

Advertising and fashion

AI-generated models, virtual influencers and synthetic imagery. Transparency obligations combine with image rights of any identifiable person and with advertising disclosure rules under the Audiovisual Media Services Code.

Enforcement: who does what

• AgID and the National Cybersecurity Agency — designated national authorities under Law 132/2025;
• Garante per la protezione dei dati personali — where AI processing involves personal data. Its record is active: a fine against OpenAI in December 2024, enforcement against an AI companion service, and a block on a foreign AI service over transparency failures;
• AGCOM — audiovisual and platform obligations;
• Criminal courts — for the dissemination offences under Law 132/2025.

AI Act penalties run to substantial percentages of global annual turnover for the most serious infringements, with lower tiers for other breaches. National sanctions apply in parallel.

What to do now

1. Inventory your AI use. Which tools, for what, at which stage of production, with what output.
2. Classify by risk tier. Most creative-industry use is limited-risk, but establishing that is itself part of compliance.
3. Verify training data for the tools you license. The GPAI summaries make this possible.
4. Decide your opt-out position for your own catalogue, and exercise it in machine-readable form if that is the decision.
5. Document the human creative contribution. Article 1 protection turns on it, and it is an evidential question.
6. Update contracts. AI training, AI modification, voice synthesis, posthumous use and withdrawal — as separate consents, not as part of a general grant.
7. Build labelling into the workflow rather than adding it at delivery.
8. For tax credit productions, get the AI clause into the templates before the first agreement is signed.

Frequently asked questions

Is AI-generated content protected by copyright in Italy?

Not where it is purely AI-generated. Law 132/2025 amended Article 1 of the Copyright Act to protect works created with AI assistance where the human creative contribution is genuine. Prompt engineering alone does not establish authorship, and unprotected output can be used by anyone.

What does Italian Law 132/2025 add to the EU AI Act?

The amendment to Article 1 of the Copyright Act, criminal sanctions of one to five years for unlawful dissemination of AI-generated content causing unjust harm, national provisions on vocal and visual identity, express regulation of text and data mining with a rightsholder opt-out, sector rules, and the designation of national authorities. Italy is the first EU member state with comprehensive national AI legislation.

Do I have to label AI-generated content?

Yes, under Article 50 of the AI Act: AI-generated content must be marked in a machine-readable format, and deepfakes resembling real persons, places or events must be disclosed. Where the content forms part of an evidently artistic or fictional work, the obligation is limited to disclosure that does not hamper display or enjoyment — it governs how you disclose, not whether.

Can I use an artist’s voice with AI?

Not commercially without that artist’s express consent. A voice is protected as an attribute of personality independently of copyright, and Law 132/2025 adds national provisions on vocal identity. A licence on the composition says nothing about the right to imitate a singer.

Can AI be trained on my catalogue without permission?

It depends on whether you have exercised the opt-out. Text and data mining for commercial purposes is permitted unless the rightsholder has reserved the right in a machine-readable form. A catalogue that has taken no position is, in practice, available.

What does the Italian tax credit require on AI?

A mandatory clause in production contracts. Article 7(6) of D.I. 225/2024 for the national production credit, and Article 9(4) of D.I. 329/2024 for the foreign producer credit — the latter on pain of the application being inadmissible. Contracts must allow authors and performers to withhold consent to AI exploitation of their work, image or performance.

When do the AI Act obligations apply?

Prohibited practices since 2 February 2025, general-purpose AI model obligations since 2 August 2025, and the bulk of the remaining requirements including high-risk provisions from 2 August 2026.

Does an old contract cover AI uses?

Almost never. Rights in forms of exploitation that did not exist at the time of contracting are not presumed transferred under Italian law, and the silence is not construed in favour of the party seeking to rely on the grant. Pre-2023 agreements should be reviewed rather than assumed to cover AI.

How DANDI supports creative businesses on AI

We advise Italian and international clients on the intersection of artificial intelligence and creative rights:

• AI compliance audits — inventory, risk classification, gap analysis against the AI Act and Law 132/2025
• Contract clauses — AI training, modification, voice synthesis, posthumous use, withdrawal, as separate consents
• Tax credit AI clauses under both decrees, for national and foreign productions
• Copyright analysis of AI-assisted work under the amended Article 1, including documentation of human creative contribution
• Opt-out strategy for catalogues, and training data verification for tool procurement
• Voice and likeness protection, and enforcement against unauthorised AI imitation
• Transparency and labelling workflows
• Regulatory proceedings before the Garante, AGCOM and the national AI authorities
• Training for production teams and in-house counsel

We work in English, Italian and French, and are regularly instructed by foreign law firms on the Italian position within a wider matter.

Book an initial consultation with Avv. Claudia Roggero.

Related guides

Primary sources

Regulation (EU) 2024/1689 (AI Act); Italian Law 132/2025; Italian Copyright Act (Law 633/1941), as amended, in particular Articles 1, 10, 20–24, 44 and 96–97; Directive (EU) 2019/790 and D.Lgs. 177/2021; D.I. MiC-MEF 225/2024 and 329/2024; Regulation (EU) 2016/679 (GDPR).

This guide states the position at the date above. AI regulation is developing rapidly; confirm current provisions before relying on it for a specific matter.

Dandi Law Firm provides legal assistance in several Practice Areas. Check out our Services or contact Us!