Regulation
xAI loses bid to block California’s AI training-data disclosure law
California’s AB 2013 is a simple idea with sharp edges: if you sell an AI model in the state, you should publicly describe the dataset sources you trained on, when you collected them, whether collection is ongoing, and whether the training corpus includes copyrighted material or personal data. Ars Technica reports that xAI sought a preliminary injunction, arguing that dataset sources, sizes and cleaning methods are trade secrets — but the judge denied the request, saying xAI didn’t show concrete, irreparable harm (too many abstractions, too few specifics). The practical consequence isn’t the courtroom drama — it’s the new baseline: “training data transparency” is becoming a compliance surface, not a voluntary blog post.
Source: Ars Technica