AI disclosure requirements in France: what applies since 2 August 2026
The EU AI Act's transparency obligations are a Union-wide regulation, so the substance is the same in France as everywhere in the EU: from 2 August 2026, AI systems that interact with people must disclose they're AI, synthetic media must carry machine-readable marking, and deepfakes and certain AI-generated public-interest text must be labeled. What differs country by country is who enforces it, in what language your users encounter it, and how quickly complaints turn into questions.
Enforcement in France
France had not enacted its Article 70 authority designations at the time of review. A government scheme published on 9 September 2025 proposes a sectoral model: the DGCCRF (the consumer-protection and fraud-control directorate) as overall coordinator and single point of contact, the CNIL as the de-facto lead wherever AI processes personal data or biometrics, Arcom for audiovisual media and deepfakes, plus ACPR (finance) and ANSSI (cybersecurity) in their sectors, but the scheme still awaits parliamentary adoption, so the supervisory map is proposed, not final. Article 50 applies from 2 August 2026 regardless, with penalties up to €15M or 3% of worldwide turnover, and the CNIL has already published AI Act Q&A and guidance while running a voluntary AI support programme.
The four duties, as they show up on your site
- Article 50(1): chatbot disclosure
- Any AI system interacting with visitors must make that clear no later than the first interaction, in a clear and distinguishable way.A line in your terms doesn't satisfy it.
- Article 50(2): machine-readable marking
- AI-generated audio, images, video, and text must be marked in a machine-readable, detectable format.For generative systems already on the market before 2 Aug 2026, this specific duty has a transition until 2 Dec 2026; the rest does not.
- Article 50(3): emotion recognition & biometric categorisation
- People exposed must be informed the system is operating.
- Article 50(4): deepfake and AI-text labels
- Visible disclosure for deepfakes, and for AI-generated or manipulated text published to inform the public on matters of public interest.
France-specific practicalities
- The DGCCRF investigates misleading commercial practices today, an AI chatbot presenting as human support sits squarely in its existing playbook even before AI Act designations land
- If your assistant touches personal data (it almost certainly does), the CNIL is already competent under the GDPR, its AI guidance is the closest thing France has to official expectations right now
- Deepfake and synthetic-media duties (Art. 50(4)) map to Arcom's proposed remit, publishers serving French audiences should watch its guidance specifically
- Keep the French-language first-interaction screenshot in your evidence file; a disclosure only in English on a .fr storefront is a weak notice
Disclosure language should match your user-facing language. For a site serving France in French, a first-message line such as "Vous discutez avec un assistant IA" (you're chatting with an AI assistant) is the pattern to aim for, visible at first interaction, not tucked into mentions légales.
Common questions
France hasn't designated its authorities, does Article 50 still apply to us?
Yes. The regulation is directly applicable EU law: the duties bind from 2 August 2026 whether or not France has finished its national designation law. The proposed scheme (DGCCRF coordinating, CNIL, Arcom and sector regulators in their domains) tells you who is likely to knock, not whether the duty exists.
We already answer to the CNIL for GDPR, is that enough?
It helps but it isn't the same thing. GDPR compliance covers personal-data processing; Article 50 adds transparency duties about the AI itself: telling users they're talking to an AI, marking AI-generated media, labelling deepfakes. The CNIL's AI guidance is a good baseline, and evidence you'd show the CNIL (screenshots, config records) is the same evidence that proves Article 50 disclosures were live.