A company that develops or uses artificial intelligence does not automatically hold all the rights in the various parts of its project. The code may have been written by a service provider, the model supplied under licence, the corpus built from third-party content, and the name of the service already used by a competitor.
The starting point is therefore to identify the assets that make up the value of the project: they determine which protections to seek, which contracts to secure and which checks to run before launch.
Are you developing or operating an AI solution? Our team, specialising in intellectual property, helps you protect your assets, secure the rights and anticipate the risks attached to their use.
Key points
- An AI project brings together several assets. The software, the data, the model, the know-how, the name and the interface do not fall under the same protections.
- Protection and freedom to operate are two separate questions. A company may obtain a right over certain parts of its project while still infringing a third party’s rights.
- Paying a service provider, or the permission granted by a tool’s terms of use, is not enough to establish that the company holds all the rights it needs.
- These checks must happen early enough. A public presentation can compromise the patentability of an innovation. A prior rights search carried out after launch may force the company to abandon a name or a visual already rolled out.
What intellectual property questions does an AI project raise?
Protection, ownership and freedom to operate must be distinguished. An asset may be protectable without the company owning it, or belong to the company without being freely usable.
| Question to settle | What to check | Risk if overlooked |
|---|---|---|
| What can be protected? | Code, database, model, trade mark, interface, invention | No exclusive right, or unsuitable protection |
| Who owns the rights? | Employees, service providers, partners, licences | Asset used without sufficient title |
| Can it be used freely? | Licences, earlier rights, terms of use, third-party data | Infringement, contractual blockage or a change of product |
| How is evidence kept? | Contracts, versions, filings, licences, documentation | Difficulty defending or monetising the asset |
| When should you act? | Patentability, confidentiality, prior rights searches | Loss of novelty, or launch under an unavailable sign |
The firm’s note
“In our view this is a break in scale, but a continuity in the legal foundations that can be relied on.”
— Vanessa Bouchara, founding lawyer of Cabinet Bouchara & Avocats, in an interview with Carnets du Luxe.
How do you protect an AI solution and its various components?
There is no single form of legal protection for an AI solution. The code, the data, the model, the name and the interface must each be examined separately, because each may fall under a different regime.
| Asset | Possible protection | Key condition or decision |
|---|---|---|
| Source code | Copyright in the software | Check originality and ownership |
| Structured corpus | Copyright in the structure, and/or database producer’s right | Identify the investment and the rights in the content |
| Model, weights, parameters and methods | Trade secrets, contractual confidentiality and other protections to be examined according to the characteristics of the asset | Maintain effective confidentiality |
| Technical application of the AI | Patent | Demonstrate a technical solution to a technical problem |
| Name of the service | Trade mark | Check validity and earlier rights |
| Interface, icon or appearance | Copyright and designs | Examine originality, novelty and individual character |
| Generated content | Protection varies with the output | Identify the human input and the rights in the inputs |
Vanessa Bouchara’s perspective
“What we look at first is not only what can be protected, but what might tomorrow prevent the company from operating, assigning or monetising its solution.”
Vanessa Bouchara, lawyer specialising in intellectual property
Source code and software
The code of an AI solution may benefit from copyright where it is original. That right protects the way the program was written, but it does not allow anyone to appropriate an idea, a feature or the very principle of an algorithm (article L.112-2 of the French Intellectual Property Code).
For software created by an employee in the course of their duties, the economic rights are in principle vested in the employer (article L.113-9 of the French Intellectual Property Code). For an external provider, the contract must provide for a sufficiently precise assignment of the rights needed to exploit the software.
Open source components must be inventoried: their licence may impose certain obligations, in particular attribution or making available the source code of certain derivative works.
The model and the know-how
The architecture of the model, its parameters or its training methods do not necessarily attract an exclusive intellectual property right as such. Depending on the project, their protection may rest on trade secrets and on confidentiality obligations within the company and with its partners.
The company must limit access and keep a record of the files sent to employees, providers and partners. Marking a file “confidential” is not enough if it circulates freely. Article L.151-1 of the French Commercial Code.
Can an AI-based invention be patented?
Yes, if the invention provides a technical solution to a technical problem. Merely implementing an algorithm, considered independently of any technical application, is not enough to make an invention patentable. The application must also be new and inventive, hence the value of examining its patentability before any public presentation.
The INPI and the EPO both apply that technical criterion.
Where the method cannot be deduced from the product or service placed on the market, secrecy can sometimes suit better than a patent. Conversely, a process that is easy to observe or reproduce will be harder to keep confidential.
Data and the corpus
A structured corpus may, depending on the circumstances, benefit from the database producer’s right where obtaining, verifying or presenting its content required substantial investment. The producer can then object to certain extractions or re-uses (articles L.341-1 et seq. of the French Intellectual Property Code).
Protection of the database does not, however, confer rights over each of the items it contains. The corpus and the works it contains must be examined separately.
The name and the interface
The name of a service can be protected by a trade mark where it is distinctive and available. A name suggested by an AI must undergo the same prior rights searches.
The interface, the icons and certain animated elements may fall under copyright or design law.
Regulation (EU) 2024/2822 modernised European Union design law, expressly bringing in elements such as animations, movements and transitions, as well as graphical interfaces within the scope of the products concerned. Its main provisions have applied since 1 May 2025; the new European design framework has been fully applicable since 1 July 2026 (EUIPO, first phase and second phase).
An example: a start-up develops a business assistant from code designed in-house, a model supplied under licence and a corpus built up over several years. It will need to check ownership of the code, the limits of the model licence, the conditions for protecting the corpus and the measures in place to keep its own parameters or methods confidential.
Who owns the rights in the developments and the data?
Having funded a development does not necessarily mean holding all the rights in it. The risk generally arises where an employee, a provider or a partner contributes to the code, the model or the data.
The contract must first identify what each party brings and what the company will actually be able to use. It must also settle what happens to improvements made during and after the collaboration.
Terms of use may authorise the use of an output without guaranteeing exclusivity, protection or the absence of third-party rights. They may also provide for prompts to be retained. Before sending data or strategic know-how to a tool, the company must check the conditions under which the provider retains, re-uses and, more broadly, processes that material.
An example: a company asks a provider to adapt a model using its own data. The contract must state who will be able to exploit the fine-tuned model, re-use the resulting parameters or continue development after the engagement ends. Paying for the work settles none of these questions on its own.
Which rights should be checked before integrating an external model, code or data?
A component that is technically accessible is not necessarily usable in a commercial offering. The licences applicable to the code, the model, the weights, the data and the documentation must each be checked separately.
That review must cover commercial use, modification, redistribution, sub-licensing, attribution obligations and any restrictions on use.
A common difficulty is that the various components of a single solution are not necessarily under the same licence: the code may be open while the weights or the data remain subject to different conditions.
Being labelled “open source” therefore never removes the need to check which licences actually apply and on what terms.
How does a company prove it owns the assets of its AI solution?
A filing does not make up for an incomplete chain of rights. The company must first keep the development and assignment contracts, the history of contributions, the versions of the code, the inventory of licences and the documents tracing how the corpus was built.
An e-Soleau envelope can, in particular, establish a certain date of filing and preserve the description of a version of the software, of technical documentation or of certain parts of the project. It does not, however, confer any exclusive right on the person filing it. For trade secrets, the company must also be able to demonstrate the access restrictions and the confidentiality measures applied.
These records matter particularly during a fundraising round, a partnership or an acquisition, when the company has to show that it genuinely controls the assets it is valuing.
Can AI-generated content be protected and used?
A platform may authorise commercial use of an output without guaranteeing its exclusivity or the absence of third-party rights. Whether it is protected by copyright depends on the human creative choices perceptible in the result, not on the number of prompts written. A distinction must therefore be drawn between the permission given by the tool, the protection of the output and the freedom to use it.
To go further into these questions (protection of generated content, human contribution and the training of models), see our analysis of AI and copyright.
Can an output infringe a third party’s rights?
Yes, even where the user did not know the origin of the material reproduced. Before publishing or marketing an output, you therefore need to check whether it reproduces something already protected: a work, a trade mark, a design, code or the content of a database.
| Element reproduced | Main risk | Point to check |
|---|---|---|
| Text, image, music or character | Copyright | Reproduction of recognisable original elements |
| Logo, name or sign | Trade mark law | Earlier rights and likelihood of confusion |
| Shape, packaging or interface | Designs | Protected appearance and overall impression |
| Code | Copyright and licences | Origin of the code and licence obligations |
| Content of a database | Sui generis right of the database producer | Substantial extraction or re-use in the light of the producer’s rights |
| Confidential information | Trade secrets | Secrecy, commercial value and the protection measures in place |
| Identifiable universe or economic value | Free-riding | Appropriation of a third party’s investment or economic value |
Images generated in a style close to that of Studio Ghibli are a good example of this overlap of intellectual property rights.
A general resemblance to a style is not, on its own, enough to establish infringement: the elements actually reproduced must be identified, such as a character, a composition, a logo or a recognisable graphic combination.
Depending on the output, the analysis may fall under copyright, trade mark law or free-riding.
Vanessa Bouchara has published an analysis of the subject in Lexbase.
Which uses of AI should a company have cleared?
As soon as AI plays a part in a strategic asset (a trade mark, a product, software, a campaign or an innovation), the checks must be carried out before it is released or launched.
| Use | Level of review |
|---|---|
| Draft or internal research | Approved tool, confidentiality and no sensitive data that must not be given to the tool |
| Published but non-strategic content | Human proofreading and a check on third-party rights |
| Logo, campaign or interface | Terms of use, contracts, availability and rights in the creation |
| Development of a solution or a model | Mapping of assets, ownership, confidentiality and protection strategy |
| Technical innovation involving AI | Patentability analysis before any disclosure |
| Content close to a well-known work or trade mark | Legal clearance before release |
Cabinet Bouchara & Avocats acts upstream to check contracts and licences, attribute each asset to its owner and choose the appropriate protections. In the event of a dispute, the firm identifies the rights that can be relied on and gathers the necessary evidence.
Before any communication: an invention must not be made public before its patentability has been analysed. Likewise, trade secret protection requires the information to meet the legal conditions and, in particular, to be covered by reasonable measures to preserve its confidentiality.
Are you developing an AI solution or preparing its launch? Cabinet Bouchara & Avocats helps you secure your rights and protect the assets of your AI solution.
FAQ
How do you value AI-related assets in a fundraising round, a partnership or an acquisition?
In a fundraising round or an acquisition, investors first seek to check that the company genuinely controls the assets presented as its own. An imprecise assignment, a restrictive licence or insufficiently protected know-how can reduce the valuation or lead to a warranty being sought, a price adjustment, or a requirement to put things right beforehand.
Who owns the rights in an AI solution developed with a partner?
Developing a solution with a partner does not say who will then be able to use, improve, license or assign it. The contract must distinguish the elements held before the partnership from the developments carried out jointly. It must also organise the rights in the code, the data, the fine-tuning, the parameters, the documentation and future improvements. The scope for use, modification, sub-licensing and continuing the project after the partnership ends should be settled from the outset.
Can an open source AI model be integrated into a commercial solution?
Not without checking the applicable licence terms. Being labelled “open source” or “open weights” means neither that the model is free of any obligation nor that it can always be built into a commercial offering.
The licences applicable to the code, the model weights, the data and the documentation must each be examined separately. Some require attribution, restrict redistribution or set specific conditions for certain uses or forms of distribution. These terms must be checked before the model is integrated, particularly where a fundraising round, a licence or an assignment of the solution is contemplated.
Sources
- French Intellectual Property Code — Légifrance: articles L.112-2, L.113-9, L.122-5-3, L.341-1 and L.342-1.
- French Commercial Code — Légifrance: article L.151-1 on trade secrets.
- European Patent Office — 2026 Guidelines for Examination, in particular the provisions on artificial intelligence, machine learning and patentability.
- Regulation (EU) 2024/2822 of the European Parliament and of the Council on the protection of European Union designs;
- EUIPO, reform of European Union design law;
- INPI, e-Soleau service.
- Vanessa Bouchara, interview with Carnets du Luxe;
- Vanessa Bouchara, analysis published in Lexbase on Studio Ghibli.
