There was a significant ruling in the Doe v. Github case involving training of AI code generator models on open source. The United States Court of Appeals for The Ninth Circuit affirmed the district court ruling dismissing part of an action brought under the Digital Millennium Copyright Act (“DMCA”) by programmers against GitHub, Inc., and other defendants. The issues centered on whether defendants removed copyright management information (CMI) from content used to train the AI code generator models. The Ninth Circuit Stated: "We agree with the district court that plaintiffs do not state a claim under the DMCA because their allegations show that Copilot and Codex do not “remove or alter” copyright management information (CMI) from a copy of an existing protected work but instead create new works that never contained that information. See 17 U.S.C. § 1202(b)(1). We therefore affirm." It added that it declines plaintiffs’ invitation to "transform run-of-the-mill copyright infringement claims into DMCA claims." This is a very significant ruling. Had it gone the other way, it would have opened the floodgates to a wave of litigation on this issue.
This seems to be a clear application of the law to the Court's factual determination that the defendants did not remove CMI from a COPY of an existing protected work but instead created NEW works that never contained that information. If that finding is correct, then this determination in this case likely compels this result. But, under different facts, the Court might reach a different result.
It is notable that the Court only addressed Plaintiff's "output theory," meaning they claimed that the output of the AI was a copy that removed the CMI. But they lost because they didn't prove the output was a copy. The panel declined to consider, as forfeited, plaintiffs’ “input” theory of DMCA liability. So, this leaves open the possibility that different facts in another case may lead to a different result. For example, if a plaintiff can show that, based on how the AI worked, the output was a copy or if they show that the input was a copy from which CMI was removed, the result may differ. Thus, I wouldn't yet conclude that removal of CMI doesn't apply to AI as a matter of law. The facts matter.