Meta suffered a major legal setback, with U.S. District Judge Vince Chhabria allowing an AI copyright against the company to proceed, although he did dismiss some elements.
Meta is being sued over its use of copyrighted material in training its AI. In the course of the trial, it was revealed that Meta pirated at least 80 terabytes of books, removing copyright management information in the process, with the plaintiffs pointing out individuals who pirated far less have deal with steep legal consequences.
“Vastly smaller acts of data piracy—just .008% of the amount of copyrighted works Meta pirated—have resulted in Judges referring the conduct to the U.S. Attorneys’ office for criminal investigation.”
After hearing Meta’s argument to dismiss the case, Judge Chhabria has dismissed some elements, but is letting the case as a whole proceed.
The plaintiffs have alleged a sufficient injury for Article III standing. Although “the specific right created by the DMCA may be comparatively new,” Meta’s removal of copyright management information is an interference with a property right that is closely related to the “kind of property-based harms traditionally actionable in copyright.”
Moreover, the plaintiffs allege that the CMI removal facilitated and concealed actual infringement of their copyrights. Copyright infringement is obviously a concrete injury sufficient for standing. So even if the CMI removal isn’t sufficiently concrete, the plaintiffs have alleged a different injury that is traceable to the CMI removal.
But the plaintiffs have adequately alleged that Meta intentionally removed CMI to conceal copyright infringement. The plaintiffs allege that Meta “knew that Llama was especially ‘prone’ to memorizing and generating outputs of CMI unless CMI was removed from” its training data. They also allege that Meta took a number of other steps to reduce the likelihood that Llama would generate outputs that would reveal or indicate that copyrighted material was included in training datasets. Taken together, these allegations raise a “reasonable, if not particularly strong, inference” that Meta removed CMI to try to prevent Llama from outputting CMI and thus revealing that it was trained on copyrighted material. Friedman v. Live Nation Merchandise, Inc., 833 F.3d 1180, 1189 (9th Cir. 2016). This use of copyrighted material is clearly an identifiable (alleged) infringement. See Stevens v. Corelogic, Inc., 899 F.3d 666, 673– 75 (9th Cir. 2018). And that Meta voluntarily revealed that an early Llama model was trained on a particular dataset containing copyrighted material doesn’t mean that it would have had no reason to hide the use of other such datasets to train later models.
Judge Chhabria did dismiss the plaintiffs’ California Comprehensive Computer Data Access and Fraud Act (CDAFA) claims, pointing that Meta did not access the plaintiffs’ computers, which would have been required for the CDAFA to be in effect.
The CDAFA claim is dismissed without leave to amend because it is preempted. The plaintiffs do not allege that Meta accessed their computers or servers—only their data (in the form of their books). And the only reason that the books Meta acquired could plausibly be considered the plaintiffs’ data is because the plaintiffs own the copyrights on those books.
Overall, Judge Chhabria’s ruling could have serious repercussions for Meta, as well as every other company that has similarly used copyrighted material to train AI models. Meta’s legal footing seems especially precarious given the preponderance of company emails proving that executives knew the company was pirating copyrighted material.
from WebProNews https://ift.tt/2l1SyjZ
No comments:
Post a Comment