Intellectual property and AI-generated content: the state of French law (2026)

As of 17 February 2026, the French and European intellectual property framework for AI-generated content has become clearer. Two pillars guide compliance: first, French copyright requires an original human contribution for a work to be protected; second, the European AI Regulation (AI Act) imposes transparency obligations, compliance with rights holders' reservations (opt-outs) and publication of a training-data summary for general-purpose AI models. This analysis provides an operational overview and proposes immediate action for businesses, publishers and creators.

1) What French copyright law says today

Originality and human involvement are required

Under French law, copyright protects any intellectual work bearing the imprint of its human author's personality (CPI, art. L.112-1). As things stand, a “work” produced exclusively by AI, without human creative direction or decision-making, does not meet this originality requirement. It therefore receives no copyright protection. See the French Intellectual Property Code on Legifrance and the overview of these principles from INPI.

When can an AI-assisted creation be protected?

Where a natural person makes free creative choices and determines the output (selection, settings, iterations, substantial editing), the final work may qualify for copyright protection in that person's favour, provided the human creative process can be demonstrated (brief, prompts, iterations, post-processing). AI has no legal personality and cannot be a co-author.

• Good evidence practice: record prompts, versions, screenshots, source files and editing logs.
• Within businesses: organise traceability in your “AI-first” processes (see Discover the Initial journey).

Related rights, databases and the press

Independently of copyright, substantial extraction and reuse of protected databases may give rise to liability under the database producer's sui generis right. Press content also benefits from a specific related right for publishers and news agencies. Training and generation uses should be assessed against these regimes using Legifrance and the practical resources on Service Public Pro.

2) Text and data mining (TDM) and AI training

TDM exceptions and the opt-out mechanism

Directive (EU) 2019/790 (copyright in the Digital Single Market) introduced text and data mining exceptions, transposed into French law. For uses other than research, rights holders may object to TDM through an appropriately expressed opt-out (for example in metadata or a robots file, in line with good practice). AI providers must respect these reservations or risk infringing rights. Useful references: Legifrance and public summaries on Vie-publique.

Good practice for providers and integrators

• Implement state-of-the-art technical detection mechanisms to identify and honour opt-outs in source content.
• Document datasets (provenance, licences, reservations, usage quotas, proportion of protected content).
• Ensure source traceability and retain collection and filtering logs.

3) AI Act: transparency, respect for rights and implementation timeline

The European AI Regulation (AI Act) regulates AI systems according to their risk level and imposes specific requirements on general-purpose AI (GPAI) models. See the text and practical guidance on EUR-Lex, entreprises.gouv.fr and Vie-publique.

Key IP obligations

• Respect for intellectual property rights and TDM opt-outs expressed by rights holders.
• Publication of a sufficiently detailed summary of the content used to train GPAI models, using the template provided for in the regulations.
• Implementation of state-of-the-art technical measures to identify and respect rights holders' reservations.

Transparency of generated content

The AI Act provides for transparency obligations towards users and the public, including information when a person interacts with an AI system, labelling synthetic content (deepfakes) and making detection or marking tools available (often associated with Article 50 in public presentations). References: EUR-Lex, CNIL.

Timeline

The AI Act entered into force in 2024 with phased application. GPAI transparency requirements and enhanced oversight reach full implementation in summer 2026. From that date (2 August 2026), providers failing to comply with documentation and opt-out obligations face penalties. See the milestones on entreprises.gouv.fr and EUR-Lex.

4) Risk situations in 2026

• Training on cultural corpora subject to a TDM opt-out without effective filtering.
• Generating images “in the style of” a living author, creating a risk of rights infringement (or free-riding/unfair competition).
• Substantial reuse of databases (extraction/reuse) without a licence.
• Potential trademark and design infringements in outputs (simulated logos, packaging).
• No marking/declaration of synthetic content in sensitive contexts (advertising, news, finance), creating regulatory and reputational risks.
• Processing personal data for training without a legal basis or information, contrary to the GDPR and CNIL recommendations.

Public authorities have highlighted public vigilance and strong expectations of transparency in AI use: see the AI Council consultation on numerique.gouv.fr.

5) Action by stakeholder group

Model providers and integrators

• Map datasets, distinguish freely usable/licensed/protected content, and publish the required training summary.
• Implement opt-out compliance (reading metadata, robots files, sitemaps, exclusion hash lists).
• Open complaint and takedown channels for rights holders, with SLAs and logging.
• Label/detect synthetic content and disclose output limitations and risks.
• Negotiate collective/individual licences with collecting societies (e.g. SACEM, SCAM) where relevant.

Business users (marketing, product, data)

• Define generative AI usage guidelines: permitted prompts, human review, “AI-generated content” marking.
• Require contractual warranties from your provider (IP compliance chain, opt-outs, indemnification, audit).
• Arrange IP/trademark checks before release (images, slogans, sounds, code).
• Retain evidence of the human creative contribution to claim copyright protection where appropriate.

Creators and rights holders

• Express a TDM opt-out on your websites and in metadata, and monitor uses with fingerprint search tools.
• Document infringements (screenshots, hashes, URLs, dates) and send structured formal notices.
• Consider licences or framework agreements with AI businesses where economically viable.

6) Key contractual clauses to include now

• Definitions of “Training Data” and “Protected Content”; an IP and GDPR compliance obligation.
• A warranty of compliance with TDM opt-outs and the AI Act; provision of the training-data summary on request.
• A prohibition on training with customer data without written, granular authorisation.
• Indemnification for infringement of third-party rights, with negotiated caps and exclusions.
• Audit, logs and enhanced information rights in the event of an authority's investigation.
• Mandatory marking of synthetic content for the agreed use cases.

Need a robust, up-to-date template? Explore AI and law resources.

7) Operational checklist for 2026

• Identify generative AI use cases and map IP risks.
• Implement opt-out compliance and dataset documentation (AI Act requirement).
• Update contracts and usage policies, including labelling generated content.
• Negotiate the necessary licences (collective/individual); retain evidence.
• Train teams and introduce human review before publication.
• Monitor legislative and case-law developments through our AI and law analysis and CNIL guides.

We support end-to-end compliance: scoping, contracts, governance and evidence. Discover the Initial journey.

8) Additional regulatory considerations

• Data protection: legal basis, minimisation, information and a DPIA where appropriate — see CNIL.
• Commercial communications and public information: marking synthetic content and combating disinformation — see Vie-publique.
• Funding and responsible innovation: guides and support on BPI France.
• Disputes and legal remedies: practical information on Justice.fr.

Brief FAQ

Is 100% AI-generated content protected by copyright in France?

No, because it lacks human originality. A demonstrable human creative contribution is required (CPI L.112-1). References: Legifrance, INPI.

Can AI be trained on protected works available online?

The TDM exception exists, but rights holders may object through an opt-out. Effective compliance with these reservations is mandatory. References: Legifrance, EUR-Lex.

What does the AI Act require of model providers?

Transparency, a training-data summary, opt-out compliance, synthetic-content labelling and implementation of state-of-the-art measures. References: EUR-Lex, entreprises.gouv.fr.

How can an AI-assisted creation be protected?

Retain evidence of your creative direction (briefs, prompts, iterations, post-processing) and specify the rights assigned in your contracts. Reference: INPI.

What are the first steps towards compliance?

Map uses, respect opt-outs, document datasets, update contracts and label generated content. Explore our Explore AI and law resources.

In summary

French law prioritises human creation; the AI Act requires transparency and respect for rights holders. Stakeholders must structure evidence, contracts and governance. To deploy your use cases securely, Explore AI and law resources and read our other AI and law analysis.