Last year, CCIA and the Chamber of Progress submitted a joint brief to the Ninth Circuit Court of Appeals in Doe v. Github, a case centered on training artificial intelligence (AI) coding tools. This week, the Ninth Circuit published their opinion adopting our position in its ruling of the case.
This case centered on a narrow question with broad implications for AI: can Section 1202(b) of the Digital Millennium Copyright Act (DMCA), the DMCA’s copyright management information (CMI) provision, create liability when an AI tool creates code resembling copyrighted source code, and the CMI of that copyrighted code is not included?
The case hinged on whether training AI coding tools on publicly available code and outputting similar code without attribution to the original code authors violates copyright law. The joint brief argued that the answer is no, this manner of training falls under fair use, and that imposing liability under Section 1202(b) would transform a targeted anti-piracy provision into a dangerously broad and overpowering copyright standard.
The CMI provision prohibits only removing or changing copyright management information such as an author’s name, the copyright notice of an existing copyrighted work, or a copyrighted work’s licensing terms. As the joint brief argued, this provision does not create an affirmative obligation to include CMI on new works simply based on similarity to other works. We emphasized that expanding this provision beyond its initial purpose could turn it into a much broader copyright claim prone to misuse. The CMI provision in Section 1202(b), if read this way, could create significant uncertainty not only for generative AI developers but digital services and individual users who create, transform, or disseminate digital content. If every AI-generated output that resembles existing code could lead to a DMCA claim, the potential liability would greatly suppress AI development and usage.
The Ninth Circuit’s opinion rejected this overly-broad interpretation and adopted the position of CCIA and the Chamber of Progress. The opinion notes that “[t]o state a claim under section 1202, plaintiffs must therefore allege that defendants removed or altered CMI from copies of existing protected works; merely alleging that a similar or derivative work does not include the CMI, without facts showing its removal or alteration, is insufficient.”
Importantly, this decision is not a blanket defense for all AI-generated code in all circumstances, and such code can still lead to copyright or DMCA liability. However, this ruling made clear that suing based on AI code that resembles existing code just because it does not contain the watermark of the original code is not enough to establish a violation of the DMCA.
As the court recognized, this decision does not automatically immunize substantially similar works from DMCA scrutiny, and similarities between two works can still be circumstantial evidence of a removal of CMI sufficient to trigger DMCA liability. Nonetheless, the court’s decision keeps the focus of the DMCA’s section 1202(b) on prohibiting copyright infringement via altering or removing CMI from an existing copy, and maintains that creating a new work without reproducing or altering CMI from an existing copy is not prohibited under 1202(b).
