IP Law Daily, COPYRIGHT—N.D. Cal.: Digital artist must replead DMCA claims against Roblox over AI training, (Oct 7, 2026)

By Saurabh Kashyap, B.A., M.A., LL.B., LL.M.

Allegations that a gaming-platform operator stripped copyright information from 3D artwork during AI training did not plausibly establish intentional removal or distribution of the works.

A federal district court in California dismissed, with leave to amend, a digital artist’s proposed class action alleging that an online gaming and game-creation platform operator removed copyright management information (CMI) from 3D models used to train generative artificial intelligence systems. Although a machine-readable “NoAI” tag could qualify as protected CMI and use of the works for AI training could exceed applicable licenses, the artist failed to sufficiently allege that the company affirmatively and intentionally removed the information or subsequently distributed CMI-stripped copies (Beaulier v. Roblox Corp., No. 5:26-cv-02642-BLF (N.D. Cal. Oct. 2, 2026)).

3D artwork and AI training. Austin Beaulier creates 3D models and has published more than 400 original models on Sketchfab, an online repository where artists can share their works. Creators making models freely downloadable select a Creative Commons license governing downstream use, including possible attribution requirements and restrictions on commercial exploitation.

Beginning in 2023, Sketchfab also allowed creators to apply a “NoAI” tag, an HTML meta tag signaling that a work cannot be used for generative-AI data collection, dataset creation, or model training. Beaulier manually applied the tag to each of his models and alleged that it, together with other information accompanying his works, constituted CMI protected by the Digital Millennium Copyright Act (DMCA).

Roblox Corporation operates a large online gaming and game-creation platform and develops generative AI systems that can produce 3D assets. Beaulier alleged that Roblox used datasets such as Objaverse-XL, which contains references to more than 10 million 3D assets from publicly accessible repositories. He allegedly found numerous works of his within the dataset.

According to the complaint, Roblox used such references to obtain models and processed them through machine-learning pipelines involving format conversion, mesh normalization, rendering, and voxelization. Beaulier alleged that CMI was removed during that process. He asserted intentional removal of CMI under 17 U.S.C. § 1202(b)(1) and distribution of works knowing CMI had been removed under 17 U.S.C. § 1202(b)(3).

Artist had standing. Roblox argued that Beaulier lacked Article III standing because his asserted injuries were speculative and reflected generalized concerns about the 3D-asset market. The court disagreed. Beaulier alleged monetary harm and impairment of his ability to control, license, and exploit his works. Relying on Kadrey v. Meta Platforms, Inc., 2025 WL 744032, at *1 (N.D. Cal. Mar. 7, 2025), the court noted that CMI removal could constitute concrete injury.

The injury was also particularized and traceable to Roblox because Beaulier alleged that his own works were included or referenced in Objaverse-XL and that Roblox used CMI-stripped representations for AI training.

CMI removal not sufficiently alleged. The first DMCA claim failed because Beaulier did not adequately distinguish affirmative removal of CMI from a failure to preserve it. Under Stevens v. CoreLogic, Inc., 899 F.3d 666, 673-75 (9th Cir. 2018), 17 U.S.C. § 1202(b)(1) requires intentional removal and knowledge or reasonable grounds to know that the removal will induce, enable, facilitate, or conceal infringement.

Beaulier alleged that his models had CMI before entering Roblox’s preprocessing pipeline and lacked it after conversion into AI-training inputs. But that did not explain what Roblox affirmatively did to remove the information. The alleged operations instead supported an alternative explanation that CMI simply was not preserved during transformation. Under Eclectic Properties East, LLC v. Marcus & Millichap Co., 751 F.3d 990, 998-99 (9th Cir. 2014), allegations consistent with both wrongdoing and an innocent alternative did not make removal plausible.

The court distinguished Doe 1 v. GitHub, Inc., 672 F. Supp. 3d 837, 855-58 (N.D. Cal. 2023), where allegations supported an inference that programs were designed to remove CMI from reproduced output. Beaulier did not similarly allege that Roblox designed its preprocessing pipelines to eliminate CMI.

NoAI tag could constitute CMI. Roblox separately argued that the NoAI tag could not qualify as CMI because it was an HTML meta tag embedded in a webpage rather than appearing directly on the model. The court rejected that argument.

Under 17 U.S.C. § 1202(c)(6), CMI includes terms and conditions governing use of a copyrighted work. The NoAI tag allegedly communicated a condition prohibiting use of a model for generative-AI datasets or training. Beaulier also manually applied a tag to each model, sufficiently connecting the information to the corresponding work. Its intended audience being a web crawler rather than a human did not prevent the tag from qualifying as CMI.

Intent allegations fell short. The court also found insufficient allegations of intentional removal but rejected Roblox’s contention that intent necessarily required a process specifically designed to target CMI. An automated process could potentially remove CMI intentionally even if removal occurred as a side effect.

Beaulier’s own allegations nevertheless undermined his theory. Although he characterized CMI removal as an avoidable design choice, he also described it as an “inherent consequence” of using large-scale creative datasets for AI training.

He did adequately allege the separate knowledge requirement. The NoAI tag prohibited AI training, while other licensing terms allegedly restricted commercial use. Roblox’s commercial AI-training use could therefore exceed the licenses. Under Oracle USA, Inc. v. Rimini Street, Inc., 879 F.3d 948, 954 (9th Cir. 2018), a licensee may infringe copyright by exceeding the scope of its license.

Distribution not alleged. The claim under 17 U.S.C. § 1202(b)(3) independently failed. Beaulier alleged that Roblox distributed CMI-stripped works through its revenue-generating AI platform, but he provided no facts showing that his particular models or copies of them were transferred to others.

Although internal use could potentially constitute copyright infringement, it did not establish distribution. Without plausible allegations of distribution, Beaulier also could not show that Roblox knew or had reason to know that such distribution would induce, enable, facilitate, or conceal infringement.

The court dismissed both DMCA claims with leave to amend within 30 days. It noted that the parties had not had an opportunity to address the Ninth Circuit’s newly issued Doe 1 v. GitHub, Inc. decision in their briefing. Failure to cure the identified deficiencies could result in dismissal with prejudice.

The Case is No. 5:26-cv-02642-BLF.

Judge: Freeman, B.

Attorneys: Michael Anthony Acciavatti (Milberg, PLLC) for Austin Beaulier. Deeva V. Shah (Keker, Van Nest & Peters LLP) for Roblox Corp.

Companies: Roblox Corp.

MainStory: TopStory Copyright CaliforniaNews AINews GCNNews