Generating text, images, video, or music via AI has become a reflex in many SMEs and scale-ups. But as soon as this content leaves the prototype stage (website, marketing campaign, product documentation, design, training), one question always returns: **is a work created...
The generation of text, images, video, or music via AI has become a reflex in many SMEs and scale-ups. But as soon as this content leaves the prototype stage (website, marketing campaign, product documentation, design, training), one question always returns: is a work created by artificial intelligence protected, and by whom?
This subject is not theoretical. It affects the value of your assets (content, creations, brands), your contracts (clients, service providers, agencies), and your risk management (infringement, image rights, confidentiality).
In France, copyright (droit d’auteur, Intellectual Property Code) rests on a structuring idea: the author is a natural person, and the work must be original, meaning marked by creative choices specific to its author.
Legal basis: general principles of copyright in the Intellectual Property Code, notably the opening provisions on the ownership of rights and the protection of works (Legifrance, CPI).
On the EU side, the notion of originality is classically formulated as an "intellectual creation of the author's own" in CJEU case law (e.g., Infopaq).
Practical consequence: if content is generated "all by itself" by a system (without sufficient human creative input), it is difficult to qualify it as a work protected by copyright in France.
In real life, we don't just have "100% AI" or "100% human." We have gray areas. Here is a simple grid.
Situation | Example | Probability of copyright protection (FR/EU) | Business implication |
|---|---|---|---|
Automatic generation | An image produced from a very generic prompt, without retouching | Low, as human input is difficult to characterize | You risk having an asset that is difficult to defend (copying by a third party, disputes) |
AI-assisted creation | The human iterates, chooses, composes, retouches, arbitrates the final rendering | Higher if human creative choices are demonstrable | Documenting the process becomes strategic |
AI as a tool in a creative workflow | AI used as a building block (variants, cleaning, draft), then production finalized by a creative | Generally the most defensible situation | Treat AI as a tool, like Photoshop or a synthesizer, with evidence |
Key takeaway: it is not the tool that decides, it is the level and nature of human contribution.
If we are in a scenario where copyright protection is plausible, the next question is: who is the author, and who exploits it?
Under French law, the author is in principle the natural person. A company can exploit the rights if:
it has obtained an assignment (contract, clear clauses, scope, territories, duration),
or if we are under certain specific regimes (e.g., software created by an employee in the course of their duties, where economic rights are allocated to the employer according to the CPI).
For marketing content, design, visuals, and text, the operational rule remains: contractualize.
Some frequent traps when AI enters the production chain:
A freelancer "generates" AI visuals and delivers them to you without a formal assignment: you have a file, not necessarily robust exploitation rights.
Several people "co-prompt" and assemble: you can create a co-author situation or at least a governance blur.
An AI tool imposes usage conditions (license, restrictions): even if your creative contributed, the tool's contract may limit certain uses.
Even if your output is not (or not surely) protectable, it can still infringe on existing rights. This is often where disputes arise.
An AI can produce an output very close to an existing work (composition, distinctive elements, recognizable style in some cases). The risk increases when:
you explicitly ask for "in the style of…",
you work with very identifiable references,
you target universes where originality plays out in details (illustration, characters, packshots).
Even without copying a work, you can clash with a trademark (resembling logo, name too close, distinctive elements). For brand assets, a minimal availability check (similarity search) is often more useful than theoretically debating the author.
Face of a real person, voice imitation, "deepfake": the risk is both civil and reputational.
In a corporate setting, using personal data in a prompt (clients, employees, support tickets) creates a GDPR risk.
On this point, if you use free or poorly configured tools, the risk is primarily data leakage. Impulse Lab has published a useful guide on precautions to take with free tools: Free AI: Useful tools without compromising your data.
Two EU texts are already structuring many operational discussions in 2026.
The Directive (EU) 2019/790 provides for exceptions for text and data mining (TDM) with, depending on the case, the possibility for rights holders to express a reservation of rights (opt-out) in an appropriate manner.
Reference: Directive (EU) 2019/790 (EUR-Lex)
Implication: depending on the sources and the presence of an opt-out, training or data collection can be more or less contestable. For a company, this translates primarily into a supplier question: where does the training data come from, and what contractual guarantees exist?
The AI Act (European regulation on AI) establishes transparency and compliance obligations for certain systems, including requirements related to general-purpose AI (GPAI) models.
Institutional reference: information and adoption steps on the Council of the EU website.
Implication: for product and legal teams, the trend is clear, more documentation requirements, and therefore more interest in choosing providers who give usable guarantees (terms of use, data policy, opt-out options, indemnification clauses, etc.).
The goal is not to "do nothing anymore." The goal is to have a production hygiene adapted to your risk level.
You can start with a very operational classification:
Level 1 (low risk): internal drafts, summaries, non-sensitive copy variations.
Level 2 (medium risk): published content (blog, ads, landing pages), campaign visuals.
Level 3 (high risk): brand assets, packaging, widely distributed content, regulated content, voices and faces, contracts, client content.
The higher the level, the more you require: human validation, traceability, source control, contractual clauses.
If you want to be able to claim rights (or at least defend the originality of a rendering), document:
iterations (key prompts, parameters, versions),
choices (selection, composition, art direction),
retouching (source files, layers, history),
the role of each contributor.
This can be light, but constant. A well-kept project folder is better than a debate ex post facto.
For content produced with AI, your contracts (or purchase orders) should clarify:
who is responsible for prompts, sources, and validations,
the assignment or license of rights on delivered elements (when applicable),
a clause on the use of AI tools and associated constraints,
guarantees in case of infringement of third-party rights (realistic indemnification level, takedown process).
The goal is to avoid a "black hole" between production, delivery, and exploitation.
Without turning every campaign into an obstacle course, a proportionate check can include:
similarity search for names and slogans (trademarks),
checking visual elements that are too close to a known universe,
verification of image rights if a face is present,
GDPR check if data was injected.
Rather than looking for a single answer ("it's mine" / "it's free"), ask yourself these questions:
Is the asset strategic (brand, long-term reuse, massive distribution)?
Can human creative choices be described in the final rendering?
Do we have proof (versions, retouching, selection)?
Is a service provider involved, and does the contract cover assignment/license and guarantees?
Does the AI tool used give usage rights compatible with your exploitation (commercial, worldwide, exclusive or not)?
Did you explicitly ask for an artist's style, a brand, a character?
Are there identifiable persons (face, voice)?
Was internal or personal data used in the prompt or source files?
Is there a rapid takedown plan if a report arrives?
Who validates internally (marketing, product, legal) according to the risk level?
A work created by artificial intelligence is not automatically protected by copyright in France; human contribution remains central.
The main risk, in practice, is often infringement of third-party rights (works, trademarks, image rights, data).
The best strategy is a combination: light process, traceability, contracts, proportionate controls.
Is an image generated by AI protected by copyright in France? Often, protection is uncertain if the image is produced without characterizable human creative input. If a human directs, selects, composes, and retouches substantially, protection is more defensible.
Who is the author of content created with a tool like ChatGPT or an image generator? Under French law, the author is in principle a natural person. A tool is not an author. The question therefore becomes: who, among the humans involved, made sufficient creative choices on the final result.
Can one claim exclusivity on AI content? Exclusivity depends on two things: the possibility of protection (human originality) and contractual conditions (AI tool, service provider). Without these two pillars, exclusivity is fragile.
Is a prompt protected? A prompt can be protected if it is original and expresses a creation (it is not automatic). In the company, the subject is primarily contractual and regarding confidentiality: who owns and reuses the prompts, and how they are documented.
What are the most frequent legal risks for an SME? The most frequent are the involuntary reproduction of elements too close to a known work, infringement of a trademark (name, logo), and uncontrolled use of data (GDPR, trade secrets) in AI tools.
What does the AI Act change for companies using generators? For most SMEs, the indirect effect is the most important: transparency and documentation requirements on the supplier side, and rising client expectations (contracts, compliance, traceability).
At Impulse Lab, we help SMEs and scale-ups put AI into production with a value and risk-oriented approach: scoping, integration, automation, and team training.
To quickly map risks and opportunities, you can start with an audit: Strategic AI Audit: Mapping Risks and Opportunities.
To discuss your context (content, product, brand, data, suppliers), contact the team via Impulse Lab.
Our team of experts will respond promptly to understand your needs and recommend the best solution.
Got questions? We've got answers.
Leonard
Co-founder
