AI Regulation Tracker / Copyright and AI

France Studies a Bill That Would Presume AI Firms Used Protected Works

A green light on constitutionality, not a law. On March 19, 2026, France's Conseil d'Etat delivered an advisory opinion on a Senate bill that would create a legal presumption that generative AI providers used copyrighted cultural content. The court found no constitutional obstacle. The bill is still a proposal, not enacted law.

By Anthony Guerriero · The Leveraged Years Regulatory Research · Published July 21, 2026 · Last updated July 21, 2026

The short version

On March 19, 2026, the Conseil d'Etat, acting in its advisory role for Parliament, issued an opinion on a Senate proposition de loi that would establish a legal presumption that providers of generative AI systems exploited protected cultural content. The idea is to shift the burden in copyright disputes, so a rightsholder would not have to prove that a specific work was ingested; the law would presume it, leaving the AI provider to rebut. The Conseil d'Etat found no constitutional obstacle, stating that the proposed provisions do not breach fair-trial principles in civil matters. It suggested tightening the drafting, replacing "exploited" with "used," and clarifying that the presumption applies only in civil disputes and only against AI system and model providers as defined under the EU AI Act. Read this precisely. It is a non-binding advisory opinion on a bill that has not been enacted. Nothing is law yet. For US counsel advising AI developers or rightsholders, it is a strong signal that Europe is looking at reversing the evidentiary burden in AI copyright cases, and that a favorable constitutional review has cleared one obstacle to it.

Regulatory Briefing

Instrument: Conseil d'Etat advisory opinion (avis), March 19, 2026, on a Senate proposition de loi creating a presumption that generative AI providers used protected cultural content
Authority: Conseil d'Etat, in its consultative capacity to Parliament (not as a court here)
Jurisdiction: France
Bindingness: Signal, non-binding. This is an advisory opinion on a proposed bill. It has no legislative force, and the bill itself is not enacted law. A favorable opinion clears a constitutional concern; it does not create any obligation.
What Changed: A proposal to reverse the burden of proof in AI copyright cases has passed constitutional review at the Conseil d'Etat, which removes one argument against advancing it.
Who This Affects: Generative AI developers and deployers operating in France, authors, publishers and other rightsholders, and litigation and IP counsel weighing exposure and strategy.
Current Status: Advisory opinion delivered March 19, 2026. Bill remains a proposal before Parliament; not adopted, not in force.
Primary Source: Conseil d'Etat, avis on the proposition de loi on a presumption of exploitation of cultural content by AI providers, conseil-etat.fr
Last Verified: July 21, 2026

The single hardest problem for an author suing an AI company over training is evidence. The rightsholder often cannot prove that their specific book, article, or image went into the model, because the training data is opaque and the company is not eager to open it up. That evidentiary gap is where most AI copyright claims stall. The French Senate bill goes straight at it by flipping the presumption. Instead of the author proving ingestion, the law would presume that a generative AI provider used protected cultural content, and put the provider to the task of showing otherwise.

That is a serious change to how these cases would run, which is exactly why it went to the Conseil d'Etat for an advisory opinion before Parliament takes it further. On March 19, 2026, the court gave its view, and the headline is that it found no constitutional problem. In the opinion's terms, as translated, the proposed provisions "do not breach the principles of fair trial applicable in civil matters and encounter no other constitutional obstacle." A reversed presumption can coexist with a fair trial, the court reasoned, because it is rebuttable and confined to civil litigation.

What did the Conseil d'Etat actually approve?

Be careful with the word "approve." The Conseil d'Etat did not enact anything and did not endorse the policy. It reviewed the proposal for legal soundness and reported that it clears constitutional muster, while recommending changes to the drafting. It suggested replacing "exploited" with "used," to line the language up with the terminology of EU law. It recommended making explicit that the presumption operates only in civil disputes, not criminal ones, and that it runs specifically against AI system and model providers as those are defined under the EU AI Act, Regulation (EU) 2024/1689. Those are precision edits, but they matter, because they tie the French mechanism to the same defined actors and vocabulary the EU framework already uses.

How would the presumption interact with EU transparency rules?

This is the part US counsel should sit with. The EU AI Act already requires providers of general-purpose AI models to publish a sufficiently detailed summary of the content used for training. A French presumption of use dovetails with that. If a provider must disclose training-content summaries under EU law, and French law presumes use unless the provider rebuts it, the practical burden on the AI company to document and account for its training data goes up sharply. The rebuttal a provider would offer is essentially transparency about what it did and did not ingest. The two instruments push in the same direction, toward providers having to show their training provenance rather than rightsholders having to prove it.

What should US developers and rightsholders do now?

First, keep the status straight. This is an advisory opinion on a bill, not a statute. It changes no obligation today, and it would be wrong to advise a client that France now presumes infringement. What it does is remove a constitutional objection and signal legislative momentum. Second, for AI developers with any exposure in France, this is a reason to treat training-data provenance as a records problem you solve in advance. If a reversed presumption becomes law, your defense is documentation, and documentation is far cheaper to build now than to reconstruct under litigation. Third, for rightsholders and their counsel, watch this bill through the legislative process, because a rebuttable presumption of use would materially change the settlement calculus for AI copyright claims in a major EU market. The direction of travel in Europe, away from making authors prove ingestion and toward making providers account for their data, is the real signal here, and it is worth positioning for even while the law is still a proposal.

Questions professionals are asking

Is this now the law in France?

No. This is a non-binding advisory opinion from the Conseil d'Etat on a Senate bill. The bill has not been enacted. Nothing about the copyright burden of proof has changed in French law as a result of the opinion.

What would the bill actually do?

It would create a legal presumption that a generative AI provider used protected cultural content, shifting the burden of proof in civil copyright disputes. Instead of a rightsholder proving that a specific work was used in training, the law would presume use, leaving the provider to rebut it.

What did the Conseil d'Etat conclude?

That the proposal presents no constitutional obstacle and does not breach fair-trial principles in civil matters. It recommended tightening the text, replacing "exploited" with "used," confining the presumption to civil disputes, and targeting AI system and model providers as defined under the EU AI Act.

Why should US AI developers pay attention?

Because a reversed presumption in a major EU market would change litigation exposure, and because it would combine with EU AI Act training-content transparency duties to push providers toward documenting their training data. Developers with French exposure should start building provenance records now, since documentation would become the central defense if the bill becomes law.

RELATED BRIEFINGS

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any bill or requirement applies to your situation with qualified counsel in the relevant jurisdiction.