Marketing teams, technical departments and agencies produce text, images and code with generative AI every day. The questions we receive are always the same two: is that content ours? and are we at risk of publishing something that belongs to someone else? The short answer: it is yours only where there is genuine human creative work, and the risk of infringing third-party rights is real and must be managed before publication.
Who is the author of content made with AI
Since 10 October 2025 the answer has been written into Italian law. Art. 25 of Law No. 132 of 23 September 2025 rewrote Art. 1 of the Italian Copyright Law (Law 633/1941): protection covers creative works of human ingenuity, including those created with the help of artificial intelligence tools, provided they result from the author's intellectual effort.
Two consequences:
- AI is not an author: an output does not become a "work" just because it is original or of good quality;
- the tool does not exclude protection: using AI does not forfeit rights, provided the person genuinely guided and shaped the result.
This was already the approach of the Italian Supreme Court: in order No. 1107 of 2023 (the case of the digital flower RAI used as the Sanremo stage set) the Court held that using software does not in itself exclude protection, but requires a rigorous assessment of the degree of creativity of the human contribution.
What it means in practice for a company
The threshold is not set as a number of interventions: it is assessed case by case. A reasonable risk map:
| Scenario | Copyright protection |
|---|---|
| Image or text generated from a prompt and published unchanged | Highly uncertain: hard to prove the author's intellectual effort |
| Output used as a draft, then rewritten, selected and composed by a person | Likely for the part resulting from human work |
| Work designed by a person (structure, choices, style) with AI as an execution tool | Stronger, if the process is documented |
The critical case is the logo or brand asset generated on the fly: without a human creative contribution, it will be hard to claim exclusivity against imitators. For distinctive elements, real design work pays off (and, where needed, trademark registration, which follows its own rules).
The opposite risk: the output that copies
A model can return passages, lyrics or images very close to protected works. On 11 November 2025 the Munich Regional Court (GEMA v OpenAI) held that memorising song lyrics in the model and returning them at the user's request are acts of reproduction, placing liability on the model operator rather than on whoever writes the prompt. The judgment is not final and does not bind Italian courts.
For a company the lesson is operational: whoever publishes and commercially exploits content answers to rightsholders for it, however it was produced. The minimum checks before publication:
- similarity search on long texts (plagiarism checkers) and images (reverse image search);
- no "in the style of" prompts for a recognisable author, character or brand in content meant for external use;
- review of the provider's terms: who owns the output, which commercial uses are allowed, whether copyright warranties or indemnities exist and on what conditions.
Your content as training data
Then there is the other side: the text and images on your website can end up in training datasets. The Italian rule, which transposes EU Directive 2019/790:
- Art. 70-ter of the Copyright Law allows text and data mining by research organisations and cultural heritage institutions, for scientific research purposes;
- Art. 70-quater allows it for anyone, on content they have lawful access to, if the use has not been expressly reserved by the rightsholder: this is the opt-out;
- the new Art. 70-septies (introduced by Law 132/2025) clarifies that these rules also apply to mining carried out with AI models and systems, generative ones included;
- Art. 26 of the same law added letter a-ter to Art. 171: reproducing or extracting text or data from works available online or in databases in breach of Arts. 70-ter and 70-quater, including through AI systems, is a criminal offence.
On the provider side, Regulation (EU) 2024/1689 (Art. 53) has required providers of general-purpose AI models, since 2 August 2025, to adopt a policy to comply with copyright law and with rights reservations expressed under the Directive. The Code of Practice published by the Commission on 10 July 2025 commits signatories to using crawlers that respect robots.txt.
If you do not want your content to feed models, you therefore need a reservation clause in your website's terms of use and robots.txt rules for training crawlers. Mind the other direction too: if you scrape content to build your own system, you must respect other people's reservations.
The SME checklist
- Internal policy: add copyright and output rules to your company AI-use policy (who may use what, for which external content).
- Record of the human contribution: for valuable content, keep drafts, versions and design choices; they are the evidence of the author's intellectual effort.
- Contracts with agencies and freelancers: require them to disclose AI use, warrant ownership of the rights assigned and indemnify you against infringement.
- Pre-publication check on similarities and provider terms.
- Opt-out on your website, if that is a company decision: clause + robots.txt.
- Transparency: align everything with the AI Act Art. 50 obligations on generated content and deepfakes.
Where to start
The first step is knowing where AI produces content in your company, and with which tools and contracts. An independent gap assessment lines up policies, provider terms and agreements with whoever creates for you, and shows where the risk is real and where it is not.