France’s Senate chamber. Photo Credit: Jacques Paquier
Is it time to make “guilty until proven innocent” the norm in training-focused lawsuits against gen AI developers? SNEP, SACEM, and 79 other French organizations seem to believe so, as they’re aggressively backing a related bill.
Those entities, plus different music-space reps, film-sector groups, and others yet, just recently voiced support for the legislation in an open letter. And this endorsement (including a petition with over 25,000 signatures to boot) follows the measure’s unanimous early April passage in France’s Senate.
But as spelled out by the campaign participants, despite the across-the-board Senate signoff, the National Assembly still hasn’t gotten the ball rolling by adding the bill to its agenda.
It remains to be seen whether the present push will deliver the desired result for the signatory organizations – a portion of which have framed the measure as a means of bringing AI giants to the table for licensing negotiations.
In practice, it’s possible that the presumption of guilt/liability (and the broader obstacles associated with an unfavorable regulatory climate) could instead compel leading AI developers to scale back or cease operating in France altogether.
Some models have trained on authorized materials from the outset; would they therefore benefit under the law? What about the impact (or lack thereof) on AIs created by companies based in nations with less robust IP protections? And the fact that potentially infringing outputs are easy enough to share across borders?
Of course, we don’t have answers to these multifaceted questions at this stage of the game. However, assuming Bing’s translation of the French-language bill is accurate, it’s safe to predict that the measure would put significantly more pressure on AI companies in domestic courtroom confrontations involving “cultural content.”
“Unless proven otherwise, in any civil dispute, the work or subject matter protected by copyright or by a related right, within the meaning of this Code, shall be presumed to have been used by the provider of the artificial intelligence model or system, provided that an indication relating to the development or deployment of that system or to the result generated by it makes such use likely,” the translated bill reads.
On an adjacent note, multiple years and twists later, we’re now just three months out from the implementation and enforcement deadline for “the majority” of the EU’s AI Act provisions.
That includes but isn’t limited to mandatory labels, “marked in a machine-readable format,” identifying AI outputs “as artificially generated or manipulated.”
