Introduction
On 21 May 2024, France adopted the SREN Law. The SREN Law contains a wide array of provisions with the aim to inter alia protect children from online pornography, combat online fraud, regulate the cloud sector or adapt French law to the EU Digital Services Act (“DSA”) and Digital Markets Act (“DMA”).
Among the numerous provisions included in the SREN Law, two articles amend the French Criminal Code to explicitly prohibit non-consensual deep fakes with certain exceptions.
Deep fakes are fake images, videos or recordings, usually generated by artificial intelligence (“AI”), that appear authentic. Examples include the image of Pope Francis wearing a white puffer jacket, the robocalls mimicking President Biden urging people not to vote in the Democratic Primary or sexually explicit images of Taylor Swift that circulated on social media.
The French Criminal Code now explicitly prohibits non-consensual deep fakes, with certain exceptions (1.), and with more severe rules for pornographic non-consensual deep fakes (2.). Will these provisions conflict with the upcoming requirements on deep fakes in the EU Regulation laying down harmonized rules on artificial intelligence (“AI Act”)?
General prohibition of non-consensual deep fakes under French law
The SREN Law supplements Article 226-8 of the French Criminal Code to explicitly prohibit the act of sharing visual or audio content generated by algorithmic processing and representing the image or speech of a person, without their consent, unless it is obvious or expressly mentioned that the content is algorithmically generated.
Article 226-8 of the French Criminal Code already prohibited the act of sharing a “montage” made from the image or speech of a person, without their consent, unless again it is obvious or expressly disclosed that it is a montage. However, certain legal scholars argued that the notion of “montage” refers to traditional video editing techniques and does not cover the transformation of images by artificial intelligence. Accordingly, this could have raised an issue given that French criminal law must be restrictively interpreted. The SREN Law therefore explicitly aims at extending the prohibition under Article 226-8 of the French Criminal Code to deep fakes.
The new provisions create however some legal uncertainty:
- First, the notion of content “generated by algorithmic processing” is broad. It goes beyond content generated by AI strictly speaking, as it extends to images or videos that are altered by photo or video editing software that do not rely on AI.
- Second, the exception for content which is “obviously” algorithmically generated will certainly give rise to debates in concrete cases. French courts will have to clarify how this is to be assessed on a case-by-case basis.
The above offence is punishable by up to 1 year of imprisonment and/or a fine of up to 15,000 Euros. If the offence is carried out via an online public communication service (e.g., a social network), it is punishable by up to 2 years of imprisonment and/or a fine of up to 45,000 Euros. Since private correspondence does not constitute online public communications (see Article 1 of the French Law of 21st June 2004 on confidence in the digital economy, “LCEN”), the sharing of non-consensual content via email or instant messaging would be subject to the lesser penalties.
Specific prohibition of pornographic non-consensual deep fakes
France has banned the practice of revenge porn (which consists in sharing, without a person’s consent, speech or images of a sexual nature obtained without that person’s consent) since 2016. However, this prohibition of revenge porn does not extend to content that is artificially generated or altered, hence the need to introduce specific provisions for pornographic non-consensual deep fakes with a new Article 226-8-1 of the French Criminal Code.
This article prohibits the act of sharing visual or audio content of a sexual nature generated by algorithmic processing and representing the image or speech of a person, without their consent, even if it is obvious or clearly disclosed that it is a deep fake. Article 226-8-1 of the French Criminal Code also prohibits the act of sharing a montage of a sexual nature made with the image or speech of a person, without their consent.
The above offence is punishable by up to 2 years of imprisonment and/or a fine of up to 60,000 Euros. If the offence is carried out via an online public communication service, it is punishable by up to 3 years of imprisonment and/or a fine of up to 75,000 Euros.
Questions around compatibility with EU law
The SREN Law was notified to the EU Commission to assess the purported provisions in light of EU law via the TRIS procedure on several occasions in the course of the legislative process. This led to detailed opinions from the EU Commission challenging the incompatibility of certain provisions of the SREN Law with the country of origin principle and the DSA. However, the EU Commission did not officially comment on the provisions regarding deep fakes.
There will also likely be questions around the compatibility with the AI Act of the new French regime regulating non-consensual deep fakes. Recital 9 of the AI Act states that national laws limiting the use of AI systems may still apply if the subject-matter is not addressed by the AI Act or pursues other public interest objectives not pursued by the AI Act. This will be a matter of interpretation of the AI Act by both national courts and EU institutions.
