France 2026: Neighbouring Rights, AI Training and Copyright, What Publishers and Creators Must Do Now

Two legislative developments in spring 2026 have fundamentally altered the landscape of AI copyright in France, creating urgent compliance obligations for every publisher, broadcaster and content creator operating in the French market. On 26 March 2026, the Assemblée nationale adopted a proposal of law to strengthen the effectiveness of press publishers’ neighbouring rights, empowering ARCOM to impose administrative sanctions and mandating platform reporting obligations. Less than two weeks later, on 8 April 2026, the Senate adopted a separate bill creating a rebuttable presumption that AI providers have used protected works in their training processes, effectively reversing the burden of proof in civil copyright disputes.

Together, these measures give rights holders significantly stronger negotiating and litigation positions, while imposing new documentation, licensing and governance requirements on AI providers and the media companies that supply them with content.

Executive Summary, What Changed and Immediate Actions

The March–April 2026 legislative window produced two converging texts that redefine the relationship between rights holders and AI providers under French intellectual property law. Rights managers and in-house counsel should focus on four immediate priorities:

• Compliance triage. Audit every existing licence agreement that permits digital reproduction, syndication or data access. Identify clauses that may inadvertently authorise text and data mining (TDM) or AI training use. Flag contracts that lack explicit AI-use restrictions.
• Licensing negotiation. Begin or accelerate negotiations with platforms and AI providers. The new neighbouring-rights framework gives ARCOM authority to set reporting conditions and impose sanctions of up to 1% of global turnover for non-compliance, creating meaningful leverage for publishers in commercial discussions.
• Evidence and documentation. Under the Senate’s presumption text, rights holders only need to present a plausible indication that their works were used. Start compiling metadata, timestamps, watermarking records and licensing logs now, this evidence forms the foundation of any enforcement action.
• Enforcement pathways. Evaluate whether to pursue civil remedies (injunctions, damages), administrative complaints to ARCOM, or collective-management channels through organisations such as the Droits Voisins de la Presse (DVP). The new framework supports all three routes simultaneously.

The sections below provide the legal background, practical checklists and model contract language that publishers and creators need to act on these priorities.

Background, French Copyright Law and Neighbouring Rights Framework

Overview of the French Intellectual Property Code

France’s copyright regime is codified in the Code de la propriété intellectuelle (CPI), which protects original works of authorship across all media. Only a human being can be recognised as an author under French law, a principle that remains unchanged in 2026 and carries direct implications for AI-generated outputs. Neighbouring rights (droits voisins), established alongside traditional copyright, protect the investments of performers, phonogram producers, broadcasters and, since 2019, press publishers.

The neighbouring-rights provisions relevant to publishers sit within the CPI’s framework as transposed from the EU Directive on Copyright in the Digital Single Market (Directive 2019/790). These grant press publishers an exclusive right to authorise or prohibit the online reproduction and communication of their press publications by information-society service providers, including search engines, news aggregators and, increasingly, AI-powered platforms.

The 2019 Transposition and Why 2026 Changes Matter

France was among the first EU member states to transpose the Directive’s press publishers’ right, enacting it into national law in 2019. Implementation, however, proved contentious. Platforms initially resisted licence negotiations, prompting intervention by the French Competition Authority, which took enforcement action and imposed a €250 million fine on Google in 2024 for failing to comply with commitments linked to press neighbouring rights.

In 2021, the Société des Droits Voisins de la Presse (DVP) was established under the SACEM umbrella to manage collective licensing. Despite these developments, rights holders continued to face structural obstacles: opaque platform algorithms, insufficient usage data, and an evidentiary burden that made enforcement slow and expensive. The 2026 reforms directly target these gaps.

Neighbouring-Rights Bill (Adopted 26 March 2026), Practical Effects for AI Copyright in France

The proposal of law adopted by the Assemblée nationale on 26 March 2026, referenced as T.A. n° 250 in the legislative dossier, strengthens the enforcement architecture around press publishers’ neighbouring rights. It addresses the structural imbalances that limited rights holders’ ability to negotiate fair remuneration with platforms and AI-powered services.

Who Is Covered, Publishers, Agencies, Platforms

The bill applies to press publishers and press agencies whose publications are reproduced, communicated or otherwise made available online by platforms, search engines and AI-driven content services. Critically, industry observers expect the scope to capture not only traditional news aggregators but also generative AI services that ingest press content for training or output generation, extending the neighbouring-rights framework directly into the AI copyright domain.

Sanctions and Administrative Processes

The bill empowers ARCOM (France’s audiovisual and digital communications regulator) to set conditions for platform compliance, mandate the reporting of usage data, and intervene in disputes between rights holders and platforms. Key enforcement provisions include:

• Administrative fines up to 1% of global turnover for platforms that fail to comply with reporting requirements or refuse to negotiate in good faith.
• Prohibition on de-ranking, platforms may not penalise publishers in search results or news feeds as retaliation for exercising neighbouring rights.
• Arbitration powers, ARCOM can facilitate or impose dispute resolution where commercial negotiations fail.
• Mandatory data disclosure, platforms must provide the usage data necessary for calculating fair remuneration, removing a longstanding opacity barrier.

For publishers negotiating with AI providers, the practical effect is significant: the threat of administrative sanctions and mandatory data disclosure gives rights holders leverage that did not exist under the previous framework.

Presumption of AI Use, What the Senate Text Does and Its Legal Effect

On 8 April 2026, the French Senate adopted a separate proposal of law creating a rebuttable presumption that AI systems have used protected cultural works in their training and deployment processes. This text, which complements the implementation of the EU AI Act in France, represents one of the most far-reaching interventions in the AI copyright debate anywhere in Europe.

Practical Legal Consequences for AI Providers and Publishers

The mechanism operates as follows: where a rights holder presents a plausible indication that their protected work was used by an AI system, whether in training, fine-tuning or output generation, the burden shifts to the AI provider to prove that the work was not used. In civil proceedings, this reversal eliminates what has been the most significant practical obstacle for rights holders: the near-impossibility of proving exactly which works were ingested into opaque training datasets.

For publishers and creators, the presumption transforms the litigation calculus. Pre-trial discovery and evidence demands that previously fell on the claimant now rest with the AI provider, who must demonstrate clean provenance for its training corpus. Early indications suggest this will accelerate settlement negotiations, as AI providers face substantial costs in producing the documentation necessary to rebut the presumption.

Limitations and Possible Defences for AI Providers

The presumption is rebuttable, not absolute. AI providers retain the ability to disprove use by presenting training logs, corpus manifests, licensing records or technical evidence of exclusion (such as robots.txt compliance or opt-out mechanisms). The likely practical effect will be to incentivise AI providers to maintain meticulous, auditable records of training data provenance, a compliance burden that mirrors the transparency obligations already emerging under the EU AI Act.

It is important to note that the Senate text still requires reconciliation with the Assemblée nationale before final adoption. The legislative process remains open, and the final text may differ in detail from the Senate version.

Practical Compliance and Licensing Steps for Media Companies

The 2026 reforms demand a structured compliance response from publishers, broadcasters and agencies. The following roadmap breaks the work into three phases aligned with the legislative timeline and the practical realities of contract renegotiation.

Short-Term (0–3 Months), Risk Triage and Immediate Audit

• Audit existing licences. Review every content licensing, syndication and data-access agreement. Identify any clause that permits TDM, automated extraction or “research purposes”, these may be interpreted as authorising AI training.
• Establish an internal rights register. Create a centralised log of all published content, associated metadata, publication dates, authorship and any existing opt-out declarations (e.g. robots.txt, machine-readable rights reservation).
• Issue formal opt-out notices. Where the text and data mining exception under the CPI applies, ensure machine-readable opt-out notices are in place across all digital properties. The opt-out must be expressed in an appropriate manner, publication in robots.txt or in metadata headers, to be effective under Article L.122-5-3 of the CPI.
• Brief internal stakeholders. Ensure editorial, legal, commercial and technology teams understand the new rules, particularly the implications of the presumption of use for evidence preservation.

Mid-Term (3–6 Months), Licence Updates and Platform Negotiations

• Update standard publishing licences. Add explicit AI-use clauses to all new and renewed content agreements. The model clauses below provide starting language.
• Build a negotiation playbook for platforms. Use ARCOM’s new mandatory data-disclosure powers as leverage. Request full usage data before entering remuneration discussions.
• Coordinate with collective-management organisations. Engage with DVP (under SACEM) and relevant sector bodies (Alliance de la presse) to align negotiation strategies and share intelligence on platform practices.
• Prepare enforcement documentation. Begin watermarking content, logging metadata and preserving timestamped copies for potential use in proceedings.

Long-Term (6–12 Months), Data Governance and AI Provider Contracts

• Implement corpus-tracking technology. Deploy content-fingerprinting and AI-detection tools to monitor whether published content appears in AI outputs.
• Negotiate directly with AI providers. Where commercially viable, enter direct licensing agreements that specify permitted uses, audit rights and attribution requirements.
• Review data governance policies. Ensure internal data-handling processes comply with both the CPI provisions and the transparency requirements of the EU AI Act.

Model Contract Clauses, How to License Content for AI Use

The following model clauses are designed as starting points for publishers negotiating with platforms and AI providers. Each should be adapted to the specific commercial context with qualified legal counsel.

• Clause 1, AI Training Opt-Out. “The Licensee shall not use, reproduce, adapt or otherwise process the Licensed Content for the purposes of training, fine-tuning, validating or evaluating any artificial intelligence or machine learning system without the prior written consent of the Licensor.”
• Clause 2, Permitted TDM Licence with Scope and Audit Right. “Where TDM use is expressly authorised, such authorisation is limited to [specified purpose/project]. The Licensor retains the right to audit the Licensee’s TDM activities, including inspection of training logs, at reasonable intervals and upon 30 days’ written notice.”
• Clause 3, Attribution and Downstream Use Restrictions. “Any output generated by AI systems that incorporates or derives from the Licensed Content must include clear attribution to the original source. The Licensee shall not sublicence, transfer or make available training datasets containing the Licensed Content to third parties.”
• Clause 4, Indemnity and Training-Log Escrow. “The Licensee shall indemnify the Licensor against any claim arising from unauthorised AI training use. The Licensee shall maintain training logs in escrow with [designated third party] for a minimum period of [X] years following termination.”
• Clause 5, Audit and Certification. “The Licensee shall, upon reasonable request, certify in writing that the Licensed Content has not been used for AI training purposes beyond the scope expressly authorised. Failure to provide such certification within 30 days constitutes a material breach.”

Enforcement, Remedies and Litigation Playbook

Civil Remedies

Rights holders can pursue traditional civil remedies under French law, including interim injunctions (référé), damages for copyright infringement and orders for the destruction or removal of infringing datasets. The presumption of use, once enacted, will strengthen the evidentiary position of claimants at the injunction stage, where courts assess the likelihood of success on the merits.

Evidentiary strategies should include preserving metadata, deploying content watermarking, and preparing subpoena requests targeting AI providers’ training corpus documentation.

Administrative Routes

ARCOM’s expanded powers under the neighbouring-rights bill create a parallel enforcement track. Publishers can file complaints with ARCOM alleging non-compliance with reporting obligations or failure to negotiate in good faith. Administrative fines of up to 1% of global turnover provide a deterrent that complements civil litigation.

Cross-Border Enforcement, EU Rules on AI Models Marketed in Europe

A key question for media copyright in France is whether EU copyright law applies to AI models developed outside Europe but marketed within the EU. Industry observers expect the answer to be yes. The European Parliament has affirmed that EU copyright requirements should apply to all generative AI models placed on the EU market, regardless of where servers are located or where training occurred. This position aligns with France’s approach and gives French rights holders a basis for enforcement against non-EU AI providers operating in the single market.

Contract Negotiation Checklist and Model Language

Before signing or renewing any agreement with a platform or AI vendor, in-house counsel should work through the following checklist to ensure the contract adequately protects the publisher’s rights under the 2026 framework.

1. Scope definition. Does the agreement clearly define what constitutes “AI training use” and distinguish it from other permitted digital uses?
2. Opt-out mechanics. Is the publisher’s right to opt out of TDM and AI training use explicitly preserved, with a clear mechanism for exercising it?
3. Data disclosure. Does the agreement require the platform or AI provider to supply usage data sufficient for calculating fair remuneration?
4. Audit rights. Can the publisher inspect training logs, corpus manifests and related documentation at reasonable intervals?
5. Indemnification. Does the AI provider indemnify the publisher against third-party claims arising from unauthorised use of content in training?
6. Termination triggers. Is failure to comply with AI-use restrictions specified as a material breach entitling immediate termination?
7. Governing law and jurisdiction. Is French law specified as governing law, ensuring the benefit of the presumption and ARCOM framework?

Where a counterparty resists any of these provisions, the fallback position should be a time-limited, narrowly scoped pilot licence with mandatory reporting and a short-notice termination right.

Implementation Timeline and Decision Matrix

Rights holders should select their response strategy based on the nature of their content portfolio, commercial relationships and risk tolerance. The matrix below maps the key options.

Decision options for rights owners range from a passive approach (monitor only) through audit, proactive licensing, collective-management coordination and, where necessary, litigation. The optimal strategy will depend on the volume and value of the content portfolio, the number of platform relationships and the resources available for enforcement.

Conclusion, Recommended Next Steps

The spring 2026 reforms mark a turning point for AI copyright in France. Publishers and creators now have both the legal tools and the regulatory infrastructure to enforce their rights against AI providers and platforms in ways that were previously impractical. The combination of ARCOM’s administrative powers, the reversed burden of proof and the strengthened neighbouring-rights framework creates a compliance environment that demands immediate action from all market participants. Rights holders who begin their compliance audits, contract updates and enforcement preparations now will be best positioned to protect their content and negotiate fair value in an AI-driven market.

Sources

1. Assemblée nationale, Dossier législatif: Renforcer l’effectivité des droits voisins de la presse (T.A. n° 250)
2. ADAGP, Announcement on Senate adoption of presumption of AI use
3. Alliance de la presse, Statement on neighbouring rights and AI bills
4. Music Business Worldwide, Bill forcing AI firms to prove non-use of copyrighted content
5. CMS Expert Guide, AI laws and regulations in France
6. Le Fil IA, Analysis of Senate presumption and burden reversal
7. Ministry of Culture (France), Press releases and guidance on neighbouring rights
8. SACEM / DVP, Droits Voisins de la Presse