In a major loss for the nation’s music industry, the Supreme Court on Wednesday ruled that a major internet service provider is not liable for copyright infringement because it failed to kick known copyright violators off its network.
Reading through the opinion, I wouldn’t be surprised to see this ruling come up in defense of chatbots trained on copyrighted works.
A provider induces infringement if it actively encourages
infringement through specific acts. Grokster, 545 U. S., at
942 (Ginsburg, J., concurring). For example, in Grokster,
we held that a jury could find two file-sharing software com-
panies liable for inducement. Id., at 941 (majority opinion).
The companies promoted and marketed their software as a
tool to infringe copyrights. Id., at 926. The “principal ob-
ject” of their business models “was use of their software to
download copyrighted works.”
“Sure, it can rip off copyrighted works, but your honor, we pinky promise that was never our principal object”. I could see it flying. Interestingly enough, the US Solicitor General explicitly brought up DMCA safe harbor in its amicus brief (siding with Cox):
The Digital Millennium Copyright Act (DMCA),
Pub. L. No. 105-304, 112 Stat. 2860 (17 U.S.C. 512), gave
service providers, including ISPs, a safe-harbor defense
to claims of copyright infringement. That defense
shields ISPs from liability for copyright infringement
based on, among other things, “the provider’s transmit-
ting, routing, or providing connections for, material
through a system or network controlled or operated by
or for the service provider.” 17 U.S.C. 512(a). To qual-
ify for that safe harbor, the service provider must
“adopt[] and reasonably implement[] * * * a policy that
provides for the termination in appropriate circum-
stances of subscribers * * * who are repeat infringers.”
I’d expect this admin to brief the court in a way that favors Musk et al, and it kind of makes sense that you’d want to bolster safe harbor protections, but I imagine a safe harbor defense of LLMs would require the reasonable policy of not training your LLM on a bunch of copyrighted works without their permission, with the express intent of creating derivative works on demand for your paying clients.
Reading through the opinion, I wouldn’t be surprised to see this ruling come up in defense of chatbots trained on copyrighted works.
“Sure, it can rip off copyrighted works, but your honor, we pinky promise that was never our principal object”. I could see it flying. Interestingly enough, the US Solicitor General explicitly brought up DMCA safe harbor in its amicus brief (siding with Cox):
I’d expect this admin to brief the court in a way that favors Musk et al, and it kind of makes sense that you’d want to bolster safe harbor protections, but I imagine a safe harbor defense of LLMs would require the reasonable policy of not training your LLM on a bunch of copyrighted works without their permission, with the express intent of creating derivative works on demand for your paying clients.
Opinion: https://www.supremecourt.gov/opinions/25pdf/24-171_bq7d.pdf
US SG amicus brief: https://www.supremecourt.gov/DocketPDF/24/24-171/359730/20250527172556075_Cox-Sony.CVSG.pdf