The rise of generative AI raises major questions in terms of intellectual property. Understanding its definitions, measuring the issues and anticipating legal risks becomes a strategic lever to create, train and exploit AI in full compliance.
What is intellectual property?
Intellectual property brings together all the rights that protect creations: texts, images, software, databases, etc.
Two main branches
- Industrial property: protects creations of a technical or commercial nature (patents, trademarks, designs & models, geographical indications, plant varieties, semiconductor topographies, etc.)
Example: filing a patent on a method of automated medical diagnosis by AI. - Literary and artistic property: protects works of the mind (copyright, related rights, sui generis right of database producers, etc.)
Example: a novel co-written by an author and an AI, creatively reworked by the human.
Intellectual property & AI: overview of the rights concerned
In the field of AI, several branches of intellectual property may apply. Each provides protection regimes adapted to the nature of the components concerned, whether it is the model, the data or the generated contents.
1. Copyright
Copyright is the branch of intellectual property law that protects works of the mind and grants their creator a set of exclusive rights, both moral and patrimonial, over the exploitation of his work.
Example: An illustration AI trained on paintings protected by copyright can generate images reproducing recognizable elements, thus exposing to a risk of infringement. Without prior authorization, the use of these works in training or generation may constitute infringement, except for legal exceptions.
2. Sui generis right of databases
The right of the producer of databases, called sui generis right, is a specific right provided by the Intellectual Property Code that protects the substantial investment made for the constitution, verification or presentation of a database, independently of the originality of its content.
Example: A company specialized in precision agriculture collects for 10 years data from sensors, satellites and drones and trains an AI model capable of predicting crop diseases.
This data is gathered, cleaned, structured and stored in a proprietary database that required a substantial investment in financial, technical and human resources. This database is protected by the sui generis right, which covers the obtaining, verification or presentation of the data.
3. Trade secrets
Trade secret is a legal notion that protects the strategic and confidential information of a company against the obtaining, use or illicit disclosure by third parties.
Example: A start-up develops a medical AI model based on a proprietary algorithm, an exclusive dataset and an innovative preprocessing method. This information, having commercial value and protected by confidentiality measures, falls under trade secrets.
4. Trademark law
Trademark law is the branch of intellectual property that governs the creation, registration, use and protection of distinctive signs used to identify the products or services of a company and to distinguish them from those of competitors.
A trademark is a sign susceptible of graphic representation or of clear and precise description, serving to distinguish the products or services of a natural or legal person.
Example: A company launches a virtual shopping assistant and registers this name and its logo as a trademark with the EUIPO. Trademark law grants it exclusive use for the registered products and services and protects against any similar use creating confusion.
5. Patent law
Patent law is the set of legal rules that govern the granting, exploitation and protection of invention patents.
The patent is an industrial property title delivered by a competent authority (for example, the INPI in France, the European Patent Office or the USPTO in the United States) which grants its holder an exclusive right over an invention.
Example: A medtech company develops an innovative AI algorithm to detect certain cancers early from MRI images. The patent grants it a 20-year exploitation monopoly and protects against any unauthorized use. This protection can be commercially valorized via licenses granted to other actors.
Moreover, depending on the nature of a work or an innovation, different protections may apply simultaneously. Several rights may coexist on the same creation.
For example, a logo can be protected by copyright, but also by trademark law.
The different components of AI: which legal regimes apply?
With the multiplication of cases related to AI in the field of intellectual property, it is essential to understand that AI does not constitute a homogeneous entity from a legal point of view.
In reality, it is composed of several distinct elements, and it is therefore crucial for the concerned operator to identify precisely which one is at stake in order to determine which legal regime applies in terms of intellectual property.
1. The AI model/system itself
Here, the AI system or model is considered as an intangible asset, patentable technical object or digital product or component.
Ex: A company develops an AI model or system.
- Applicable rights: patent, copyright (code), trade secrets
- Intellectual property issues: patentability, protection of source code as software, economic advantage.
2. The training databases (corpus & fine-tuning)
Here, the data used to train the AI is considered, generally including text, images, audio, video, or other structured or unstructured formats.
Ex: Various data: novels and short stories, press articles, paintings, drawings, recorded pieces, digitized scores, stock market prices, annual reports, economic indices, tweets, LinkedIn posts, public conversations, etc.
- Applicable rights: copyright (on works in the corpus), sui generis right (structure/obtaining), trade secrets
- Intellectual property issues: licenses, respect of exceptions, traceability of sources, TDM opt-out
3. The outputs (results generated by AI)
Here, the result or the answer generated by the system is considered, that is to say a value representing all or part of the operation performed by the latter from the input data.
Ex: Complete script of a short film produced by ChatGPT from the author’s synopsis
- Applicable rights: copyright (if original human contribution), neighboring/specific rights depending on the country, responsibility with respect to third-party rights.
- Intellectual property issues: risk of infringement/plagiarism if the output reproduces recognizable elements; confusion/substitution.
Training data and generated contents: what are the risks in terms of intellectual property for AI users?
Risks related to training data
Datasets used to train a model may contain protected works (texts, images, music, videos). Their collection by scraping or by text and data mining (TDM) may constitute an act of reproduction requiring the authorization of rights holders.
Even if a TDM exception is provided in Europe, the opt-out mechanism allows creators to oppose the use of their works. Failure to respect these oppositions exposes developers to litigation for copyright infringement.
For AI users, this means they may be exposed to indirect legal risks if the models they use rely on data collected illegally or contested.
Risks related to generated contents
An AI can produce results that reproduce the style, composition or recognizable elements of existing works. If the nature of the generated work is not judged sufficiently transformative, it may be considered as infringement. The problem is accentuated when the generated content can be perceived as substitutable for the original work, or when it creates confusion in the public’s mind.
For AI users, this implies a risk of seeing their use of generated contents lead to accusations of infringement, including in a professional or commercial context, with financial and reputational consequences.
These risks remind us that the exploitation of AI requires increased vigilance and proactive management of intellectual property, in order to reconcile innovation and respect for creators’ rights.
Adopt operational AI governance today
The exploitation of AI requires constant vigilance and rigorous management of intellectual property. To reconcile innovation and respect for rights, theoretical compliance is not enough: it is necessary to set up operational governance
Discover the Naaia platform and schedule a demo: Access our AI management platform.