What does "AI copyright" mean?

AI copyright refers to the copyright classification of content created or edited using artificial intelligence. Purely machine-generated AI outputs generally do not enjoy reliable, independent copyright protection in the EU. An AI-supported final version may be eligible for protection if a human has demonstrably shaped the result through free and creative decisions.

Legal status: January 2026. This glossary entry provides companies in South Tyrol, Italy, and the EU with a general overview of AI copyright law. For economically significant or disputed individual cases, you should consult a legal advisor specializing in copyright law.

AI ( Artificial Intelligence) is the umbrella term for digital systems that recognize patterns in data and take over tasks that would otherwise require human perception, assessment, or decision-making... Click to learn more . AI is a tool, not an author. What matters is not the length of a prompt, but the discernible human contribution to the finished work.

AI copyright: When is an AI output protected?

EU copyright law requires original intellectual creation for protection . According to the case law of the European Court of Justice, the author's personality must be expressed through free and creative decisions . The ECJ has specified this standard of originality, among other things, in its judgment C-145/10 "Painer".

The frequently used term "level of originality" simply describes the required level of creative independence. In the context of EU law, artistic merit, beauty, effort, or working time are irrelevant. What matters is whether an identifiable human creative contribution shapes the concrete expression of the work.

Four typical cases can be distinguished for AI copyright:

• Fully autonomous output: If the system generates the concrete expression without human creative control, human authorship is regularly lacking.
• General instructions: A short PromptThe term "Prompt (AI)" may sound like technical jargon at first, but it actually encompasses an exciting world that has a lot to do with the way... Click to learn more like “Create a modern Company LogoDefinition of a Logo: A logo (also known as a brand logo, product logo, company logo, or logotype) is the graphic representation of the name of a product, company, or organization. It... Click to learn more“ sets a goal, but hardly shapes the concrete result.
• Iterative process: Multiple precise instructions, a conscious choice of options, and traceable processing can prove human decisions, but do not guarantee protection on their own.
• Substantially revised final version: If a person independently develops composition, language, dramaturgy, forms, or code, these human-created components can be protected.

In 2025, the European Parliament noted that there are no specific EU-wide rules that grant autonomously generated AI output its own copyright protection. The protectability of AI content therefore remains a case-by-case matter, linked to human creation.

AI copyright: A comprehensive prompt is not automatically sufficient

A prompt itself can be protected text if its wording achieves the necessary originality. This does not automatically mean that the AI-generated output is also protected. Between input and result lies a technical generation process, the specific output of which you often do not fully control.

For the legal assessment, it is crucial whether you actually determined the expression of the final result:

• Did you develop any specific design elements yourself?
• Were you able to deliberately control the form, structure, word choice, or composition?
• Did you simply adopt the result or did you substantially change it?
• Can you support your decisions with drafts and interim results?
• Is it possible to distinguish which components are yours and which are from the system?

A lengthy prompt can be very time-consuming without providing sufficient control over the final output. Conversely, extensive post-processing can create a unique, human-generated piece of work. It's not the number of words that matters, but rather the creative control over the final result.

Purely generated, selected, or human-designed?

Example Branding

A company has 100 logos designed and adopts one variant unchanged. Simply selecting from predefined results is often insufficient to transform the chosen logo into the user's own intellectual creation.

The assessment may differ if a designer first develops the positioning and visual concept, uses AI only for initial sketches, and then independently designs the characters, proportions, typography ( typography is the conscious design of typefaces, spacing, hierarchy, and readability in print and digital media. The English term typography means the same thing: typefaces are... Click to learn more) , color scheme, and application system. The protectable element then lies in the human-developed brand solution, not in the unprocessed AI output.

In my work with SMEs, I deliberately separate these steps: AI provides variations; positioning, selection criteria, composition, revision, and final responsibility remain with humans. I describe how this approach works in practice in the article " AI as a Tool in Brand Development ."

Example text

It is uncertain whether an AI-generated text adopted without verification enjoys its own copyright protection. However, if you develop the argumentation, structure, examples, and linguistic style yourself, while the system only provides variations of the wording, the human-created components may be protected.

AI -powered content creation should involve more than spelling corrections or replacing individual words. A fully developed editorial concept and substantial revisions more clearly demonstrate the human contribution to the design.

Example image

For a fully generated product image without any creative post-processing, it is generally not possible to reliably establish separate copyright protection. However, if a photographer develops the image, photographs essential elements themselves, and uses AI only for limited retouching, the human-influenced photograph may be protected by copyright.

Regardless of the copyright status of the publication, depicted individuals, protected designs, or third-party photographs may be affected. For company websites, you will find a separate classification regarding image rights in Italy and South Tyrol.

Example software

If you adopt an automatically generated code block without checking it, neither its protectability nor its legal compliance is guaranteed. If you develop the architecture, data model, process logic, and essential parts of the implementation yourself, these human-created components can be protected.

Software also involves licensing and security issues. AI output can reproduce code that is subject to a third-party license or is incompatible with existing components. Therefore, license verification, security testing, and technical documentation are essential components of the release process.

AI copyright in Italy

For companies in South Tyrol, Italian law is relevant in addition to EU law. Italy, with Law No. 132 of 23 September 2025, clarified that copyrighted works must be the result of human intellectual activity. Works created using AI tools are only protected if they are based on the intellectual work of the author.

The use of an AI system does not preclude protection. However, human contribution must be evident in the work and verifiable. The purely technical triggering of the system is not reliably sufficient.

The lack of in-house protection does not automatically permit free use.

An unprotected AI output must not be used automatically without review. A lack of inherent protection and permissible use are two different issues.

Despite lacking its own protection capabilities, an AI output can infringe upon the following third-party rights:

• Copyrights to texts, images, music, IllustrationsIllustration (Latin illustrare – "to illuminate, explain, praise") is a visual form of representation used to convey information, stories, or concepts in a vivid and understandable way. Illustrations... Click to learn more or software
• Trademark rights and business identifiers,
• Design rights,
• Personality rights and rights to one's own image,
• contractual confidentiality obligations,
• Trade secrets or confidential information,
• Licenses for datasets, fonts, templates, or code libraries.

If a generated design incorporates protected expressive elements of a well-known character, photograph, or existing logo, this may constitute copyright infringement or an infringement of other intellectual property rights. The label "created with AI" does not eliminate this risk.

The exam requires three separate questions: Is the publication itself protected? Are you allowed to use it according to the terms of use? Does the specific use infringe the rights of third parties?

Separate protectability, usage rights and training material

1. Protectability of the AI ​​output

This level concerns the question of whether a sufficiently human-influenced work has been created. Human authorship, originality, and the specific creative contribution are crucial.

2. Contractual usage rights

The AI ​​provider's terms of service may stipulate what rights you receive to inputs and outputs, whether content (which includes all specifically published digital content on websites, in shops, on social media channels, in newsletters, and in other digital environments) may be used for training, and what responsibilities you bear. However, contractual permission for commercial use does not prove the creation of statutory copyright or the freedom from third-party rights.

Check in particular:

• Who is allowed to use input and uploaded files?
• What usage rights are granted for the publication?
• Is commercial use permitted in the selected tariff?
• Is the provider allowed to use your content for training or to improve the service?
• Are there any disclaimers or indemnification clauses?
• Are there special rules for images, music, voices, or code?

3. Admissibility of the training material

The protectability of a result must be distinguished from the lawful processing of the training material. EU rules on text and data mining treat scientific research under Article 3 of the DSM Directive separately from general uses under Article 4. For general uses, rights holders can declare an appropriate reservation of rights, for example, machine-readable content in the case of publicly accessible online content.

The EU AI Act supplements this level for providers of general-purpose AI models. Article 53 requires, among other things, a strategy for complying with EU copyright law and a sufficiently detailed public summary of the content used for training. Transitional periods apply to models already made available before the relevant application date.

Review AI copyright within the company

For small businesses, the review should correspond to the risk of its use. The more strongly a piece of content shapes the brand ( the definition of "brand" comes from English and stands for "trademark") , is commercially exploited, or is intended for exclusive use, the more thorough the review should be.

1. Define intended use and risk

• Is the output only used internally as an idea?
• Is this about a short-lived social media post?
• Will a lasting brand motif, product, campaign, or software be created?
• Should the content be sold, licensed, or used exclusively?

An internal brainstorming session requires less validation than a company logo, packaging, or a digital product. For key brand elements, an untested AI output should not be used as the final version.

2. Document human design

• Preserve BriefingIf you want to create a briefing, you primarily need clarity: A briefing is a structured document or a coordinated conversation that outlines the goal, framework, etc. Click to learn more, concept and own sketches.
• Save relevant prompts and variants.
• Document selection decisions and their justification.
• Secure intermediate states of post-processing.
• Mark which components were created by humans or significantly altered.

An audit trail is a traceable record that documents who did what, when, whether it was changed, decided, or submitted for approval within a system. For SMEs... Click to learn more . Such documentation does not automatically create copyright, but it can demonstrate the human process behind its creation and the responsibilities involved.

3. Check the rights of third parties

• For important images, perform a reverse image search or a similarity search.
• Check names, claims and logos in relevant trademark registers.
• Check the code for known components and license information.
• Do not use confidential documents as input if the service is not authorized for that purpose.
• Obtain consent if identifiable individuals, voices, or personal data are involved.

4. Secure the provider's terms and conditions

Save the version of the terms of service that is valid at the time of production. Terms may change, and different plans for the same service may provide for different rights.

5. Define responsibility and approval

Define who controls AI spending from a technical, legal, and brand strategy perspective. An internal rule reduces the risk of employees or contractors publishing key content without verifiable review. Data protection protects the personal data of natural persons from unlawful processing, misuse, and loss of control. For SMEs, this means consciously deciding which data you collect... Click to learn more. AI regulation should be considered together; our decision logic for GDPR and the EU AI Act for SMEs offers a practical supplement.

AI copyright and labeling requirements under the EU AI Act

The EU AI Act stipulates transparency obligations for certain artificially generated or manipulated content. These include machine-readable marking of synthetic output by providers of certain systems, as well as disclosure obligations for deepfakes and certain texts that provide information on matters of public interest.

A visible labeling requirement does not apply across the board to every AI-generated output. Article 50 contains various obligations, requirements, and exceptions. For texts, human review with editorial responsibility may be relevant. Special requirements apply to artistic, creative, satirical, or fictional content.

The transparency obligations under Article 50 will apply from August 2, 2026. However, labeling and copyright address different issues: the label "AI-generated" does not determine whether a work is eligible for protection, whether it has usage rights, or whether it infringes on the rights of others.

What I recommend to SMEs regarding important brand content

AI can accelerate research, variant generation, and initial drafts. However, responsibility for brand content must remain with humans. The more long-term and economically significant a piece of content is, the more clearly the human element should be incorporated.

• For ideas: Use AI output as raw material.
• For ongoing content: Humanly verify facts, language, rights and brand fit.
• For campaigns: Develop the concept, dramaturgy, and final design in a comprehensible manner.
• For logos and brand elements: Do not accept any unverified copy; professionally verify originality and intellectual property rights.
• For products and software: Document rights, licenses, security, and the creation process.
• For commissioned work: Contractually regulate usage rights, AI deployment, liability and handover of work progress.

From my more than 20 years of experience in branding – branding is the conscious, strategic building of a brand. Branding determines how your company is perceived, what people recognize it by, and why they trust it... Click to learn more , web development, and digital strategy, the division of labor is clear: The system can provide options. Positioning, selection, editing, and responsibility must remain with humans. AI enhances existing expertise but does not replace it.

Questions and answers about AI copyright

Are purely AI-generated contents automatically in the public domain?

No. Unprotected AI output is not automatically risk-free or free from claims. Terms of use, trademark rights, privacy rights, trade secrets, and third-party copyrights may still restrict its use.

Can a prompt be protected by copyright?

A custom-written prompt can be protected as text if it itself meets the required originality standards. The protection of the prompt does not automatically extend to the AI ​​output generated from it.

How much post-processing is necessary for protection?

There is no fixed percentage and no mandatory number of editing steps. What matters is whether the final version expresses recognizable free and creative decisions made by a person. Mere corrections or minimal changes are not reliably sufficient for this.

Am I allowed to use AI-generated content commercially?

That depends on the service's terms of service and the rights involved. A commercial release from the provider does not automatically protect you from copyright infringement or other third-party claims.

Does every AI output need to be labelled?

No. The EU AI Act does not require a blanket visible label for every publication. The transparency obligations applicable from August 2, 2026, concern specific providers, deepfakes, and particular publication situations; additional national or industry-specific rules may also apply.

Who is the author when an employee uses AI?

In the EU and Italian legal framework described here, authorship is linked to the creative decisions of a natural person. Whether and to what extent the company acquires rights of use depends additionally on the applicable law, the employment relationship, and the agreements reached.

What should a contract with a contractor regulate?

The contract should stipulate the permitted use of AI, human contribution, transferred usage rights, and the handling of input, intermediate results, and confidential data. Provisions regarding third-party rights, documentation obligations, and approval processes are also advisable.

How can I prove my human authorship?

Keep concepts, sketches, versions, edit files, selected prompts, and chronologically traceable intermediate stages. Documentation doesn't automatically prove copyrightability, but it makes your design contribution more transparent.

Can AI output infringe on third-party copyrights?

Yes, if the publication incorporates protected expressive elements of an existing work. Before publishing commercially important texts, images, music, and software, check for recognizable similarities and known sources.

Does a different standard apply in Italy than in the rest of the EU?

Italy applies European copyright law within its national legal framework and, since 2025, has explicitly emphasized human intellectual contribution. For companies in South Tyrol, therefore, a verifiable human contribution remains crucial, even in AI-supported works.

Conclusion on AI copyright

In AI copyright law, it is not only the involvement of a technical tool that matters, but also who created the specific expression. A purely machine-generated output does not provide a reliable basis for exclusive copyright protection in the EU. A final version conceived, controlled, and substantially edited by a human, on the other hand, can contain protected elements.

For companies, the following applies: treat AI output as source material, document human design, review terms of use, and clarify third-party rights. Regarding the internationally debated issue of copyright for AI content, technical generation, legal protection, and contractual usage rights must be assessed separately.

This article does not constitute legal advice. For logos, product designs, extensive campaigns, software, licensing models, or a specific dispute, you should consult a specialized lawyer.

Sources

1. ECJ, Judgment C-145/10 “Painer” — curia.europa.eu (2011)
2. Copyright of AI-generated works: Approaches in the EU and beyond — europarl.europa.eu (2025)
3. Legge 23 September 2025, n. 132 — gazzettufficiale.it (2025)
4. Regulation (EU) 2024/1689, in particular Articles 50, 53 and 113 — eur-lex.europa.eu (2024)
5. Directive (EU) 2019/790, in particular Articles 3 and 4 — eur-lex.europa.eu (2019)