New French draft law on AI: Generated or not generated, that is the question

The explanatory memorandum highlights the dangers posed by AI, particularly in terms of misinformation as a result of the publication of deepfakes on social networks. This draft law also seeks to impose a transparency obligation on content produced by AI. For its drafters, it endeavours to create a clear framework in which innovation and responsibility coexist, by both protecting users and strengthening trust in digital content.

Protecting users involves easily distinguishing an artificial creation from an authentic document. The underlying idea is the fight against disinformation and manipulation.

This proposed law is also accompanied by a legal obligation for platforms to systematically implement tools capable of making such a distinction.

To give concrete expression to this objective, the sole article of this draft law requires that ‘Anyone publishing on a social network an image generated or modified by an artificial intelligence system must explicitly mention its origin’, ‘that includes a clear and visible warning specifying the use of an artificial intelligence model to create or modify the image.’ This is supplemented by an obligation on online platform operators ‘to put in place technical means to detect content generated by artificial intelligence and to check that it is labelled correctly. They must also inform their users of the obligations in force and provide a reporting tool for suspicious content’.

Thoughts

A review of the explanatory memorandum reveals that, as it was the case with the previous French AI draft law, the critical nature of AI is presented in almost a caricatural way, with the technology being depicted as a source of serious threats that need to be quickly remedied.

Furthermore, this draft law appears to wholly disregard the existence and implications of the AI Act (Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024), which merits recognition for its establishment of a tangible framework for AI [IPKat here]. In comparison, this brief draft law may be of limited relevance given the EU regulation. It is therefore open to criticism for its lack of ambition. Nevertheless, it would be futile to attempt to replicate the Regulation.

For this Kat, it would be prudent to consider the full range of challenges posed by AI, rather than focusing on a single, visible aspect. Adopting a more comprehensive approach would allow to establish a robust framework, in line with the EU regulation. For example, it is evident that AI gives rise to queries pertaining to its relationship with copyright. This is particularly evident in relation to concepts such as AI training [IPKat here], as well as the liability of AI model providers [IPKat on this point]. This Kat would therefore appreciate the delivery of a developed French draft law that takes account of European advances. Such an approach would reduce the potential risk for legislative proposals to become fragmented and sporadic, and to lack sufficient coherence when considered in isolation.