The ruling interprets Section 1202(b), which bars the removal or alteration of copyright management information (CMI) from protected material, in a way that excludes AI-generated outputs from its scope.
The lawsuit, Doe et al. v. The plaintiffs claimed this constituted a violation of Section 1202(b) by effectively stripping CMI from existing works.
The Ninth Circuit Court of Appeals dismissed the claim, upholding a lower court’s decision. The court’s reasoning centered on whether Copilot’s outputs were derived from direct copying or produced through an independent generative process. It concluded that Copilot’s method, using statistical analysis to create new code, did not involve copying and removing CMI. Instead, the AI’s training was framed as a learning process that generated original material, even if the results bore similarities to copyrighted code.
The court found that Section 1202(b) does not require literal identicality of copyrighted works to establish liability. However, it acknowledged that if Copilot functioned like a search engine-retrieving and reproducing stored code without CMI-the plaintiffs might have a stronger argument. Under current interpretation, the AI’s generative approach does not meet the DMCA’s removal standard.
The court declined to address a secondary claim: whether the defendants violated the law by removing CMI before using the code as training data, because the plaintiffs had abandoned it in earlier proceedings. This leaves unresolved whether removing CMI from copyrighted works before using them as AI training data could independently violate §1202(b).
The decision clarifies that AI-generated code resembling existing works may still infringe copyright, and the court found that the plaintiffs had not alleged removal or alteration of CMI from copies of their code under Section 1202(b).
Leave a Reply