Songwriter collecting societies from all over the world have called on French lawmakers to urgently pass a proposed new law that would reverse the burden of proof in AI copyright disputes, so that AI companies would have to prove they haven’t used specific works in their training processes when they are accused of copyright infringement.
227 societies representing songwriters, composers and other creators backed the proposal in Paris last week during the General Assembly of their global grouping CISAC. And French society SACEM has been communicating that support to French lawmakers this week.
In a statement, SACEM says that AI companies using millions of existing songs and compositions, not to mention books, newspapers, photographs, artworks and so on, all without getting permission from or compensating creators, is “the greatest plundering of creative and artistic works ever perpetrated”.
It then seeks to rally French politicians by telling them that “France invented modern copyright” and “it was also one of the birthplaces of collective rights management”. Which means, “for more than two centuries, France has been at the forefront of efforts in Europe and around the world to defend creativity, cultural diversity and the rights of creators”. And it should continue in that mission in the context of AI.
The proposed new law is contained in what is known as the Darcos Bill, named for the French senator that proposed it, Laure Darcos. It seeks to address frustration in the creative industries regarding the lack of transparency over what existing works AI companies have used when training generative AI models.
While there remains a major dispute in at least some countries over whether or not AI companies need to get permission to use existing copyright protected works when training their models, even in countries where permission is definitely needed, creators and copyright owners face the additional challenge of not knowing if their works were copied into any one training dataset.
There have been wide-ranging calls for separate transparency obligations to be introduced for AI companies. In theory, within the European Union, the EU AI Act introduced some transparency rules for AI businesses, although those rules have not yet brought about much transparency.
Given many AI models are trained with huge quantities of existing content, much of it scraped off the internet, under the Darcos Bill a copyright owner wouldn’t have to prove their work was specifically included in order pursue legal action. Instead, it would be for the AI company to prove that they hadn’t used any one person or company’s works.
The proposals were passed by the French Senate in April and are now with the National Assembly. But there have been concerns that the proposals are stalling in the lower chamber of the French Parliament.
It is due to be discussed on Thursday, although it’s the final item on the agenda for that day’s proceedings and Reporters Without Borders, which supports the bill, says that means there is “almost no chance of it being debated in the chamber”.
SACEM’s statement describes the Darcos Bill as a “straightforward response” to the growing concerns around the lack of transparency in generative AI, pointing out “it creates no new rights”, but “makes existing rights under French and European law effectively enforceable by strengthening transparency requirements for AI providers”.
Stressing that the 227 CISAC member societies together represent “millions of human creators” around the world, the statement adds that passing the Darcos Bill would “resonate far beyond France’s borders”.
It then concludes, “There is an urgent need for action on behalf of creators around the world, who know that France - as it has so often done throughout the history of copyright - can once again lead the way and send a clear message: the development of AI must not come at the expense of culture”.