Robots.txt, AIPREF, transparency, filtering of generated content… AI developers are proposing several solutions to limit conflicts with rights holders. But these mechanisms are primarily intended to prevent certain uses. They do not address the central question raised by the Darcos bill: how can one establish, before a judge, that a work was in fact used to train an artificial intelligence model? In a previous article, we noted that the submissions filed by Anthropic and the CCIA with the European Commission did not present a genuine procedural alternative to the presumption of use proposed by the Senate. However, the developers are not without proposals: their submissions put forward several avenues aimed at improving the functioning of the current framework. These include refining opt-out mechanisms, adopting…
