District court dismisses DMCA claims based on removal of copyright management information (CMI) from 3D models as part of process to train AI models, finding that uniform process to convert files into new format did not constitute “intentional” removal of CMI and that internal training did not constitute “distribution” of 3D models, but court grants leave to replead.
Plaintiff Austin Beaulier is a professional 3D artist who publishes digital models on various online repositories such as Sketchfab. Beaulier’s models were published under Creative Commons licenses, which generally allow anyone to reuse the work under certain conditions, such as providing attribution or for only noncommercial uses. When Beaulier published his models on Sketchfab, he used a feature that applied a “NoAI” tag to his works. Sketchfab therefore embedded “a machine-readable meta tag in the model’s webpage” designating his work as not for use for AI data collection or training.
According to Beaulier, Meta Platforms Inc., through an academic dataset, found his (and millions of other) models, traced them to the repositories on which they were located (including Sketchfab) and downloaded them to use for training AI systems. Beaulier claims that after downloading the models, Meta “processed them through a machine-learning system” to convert the 3D works into training inputs. This process, Beaulier claims, retained “the expressive content” of the models but removed any “attribution, licensing, and other” copyright management information (CMI). Without the CMI, the AI systems being trained could not identify and honor the designated licensing conditions attached to the 3D models.
Based on these allegations, Beaulier, on behalf of a class of creators who published 3D models under Creative Commons licenses, asserted two claims under the Digital Millennium Copyright Act (DMCA), 17 U.S.C. §1202. First, Beaulier claimed Meta intentionally removed CMI, in violation of 17 U.S.C. §1202(b)(1). Second, Beaulier claimed Meta used or distributed the works knowing that the CMI had been removed, in violation of 17 U.S.C. §1202(b)(3).
The district court granted Meta’s motion to dismiss both claims without prejudice. First, the court addressed the intentional removal claim under Section 1202(b)(1). Of the four elements required for a Section 1202(b)(1) claim, the court found that Beaulier had adequately alleged all but one. The NoAI tag and other licensing and identifying metadata constituted CMI. The CMI was removed. And Meta had grounds to know that it was removing the CMI from the work. However, the court found that Beaulier had not alleged that Meta’s removal of the CMI was “intentional” as required by the DMCA. Relying on an earlier case it had decided, the court explained that unintentional side effects of a uniform process (such as cropping photos into a thumbnail), rather than targeted acts directed at removing CMI, do not constitute intentional removal. In the instant case, per Beaulier’s allegations, Meta converted the 3D model formats to make them usable for AI training, not to strip them of CMI. Thus, the removal of all metadata was more akin to an “unintended side effect” than an intentional effort to circumvent the copyright management designations. Accordingly, the court dismissed the Section 1202(b)(1) claim.
Next, the court turned to the claim based on the alleged distribution of versions of the 3D models stripped of CMI, brought under Section 1202(b)(3). The court explained that “distribute” as it is used in the statute means “the transfer of a work to the public by sale or other transfer of ownership, or by rental, lease, or lending.” Beaulier, however, failed to allege any such transfer. He simply alleged that Meta used the CMI-stripped 3D models “within the training datasets used to develop its general artificial intelligence systems.” This type of internal use, the court found, did not satisfy the distribution element of Section 1202(b)(3). And without distribution adequately alleged, the court dismissed the Section 1202(b)(3) claim.
The court dismissed without prejudice, as its grounds for dismissal were narrow and could potentially be cured in an amended complaint.
Summary prepared by Tal Dickstein and Tyler Downing
-
合伙人 -
Associate