Could one ruling on copyright ultimately compel internet providers to disconnect millions of users and destabilize the AI industry along the way? That prospect looms over Cox Communications today, as the U.S. Supreme Court considers whether internet service providers can be held liable under contributory standards of copyright infringement for their users’ infringing activities. At the heart of the controversy are peer-to-peer protocols like BitTorrent, whose decentralized architecture makes enforcement both technically difficult and legally dicey.
BitTorrent operates by breaking up files into tiny fragments and distributing them across a “swarm” of participating nodes. Each node downloads and uploads file fragments to its peers simultaneously, with no central servers involved, and therefore, no easy way to identify a single bad actor who could be held responsible for infringing content. Enforcement typically depends on tracking IP addresses seen in a swarm, but as computer science research has demonstrated, false positives are common-printers, routers, or innocent subscribers can be mistakenly identified as infringers. This technical ambiguity feeds directly into the legal question before the Court: how much “knowledge” and “contribution” is enough to strip an ISP of safe harbor protections under the Digital Millennium Copyright Act?
The music labels, including Sony Music Entertainment, say Cox had clear evidence of “habitual offenders” and failed to act, pointing to millions of infringement notices and a termination record that saw 619,711 subscribers cut off for nonpayment but only 32 for copyright abuse. They say Cox’s policy of waiting for 13 infringement notices before action combined with internal emails like “F the DMCA!!!” demonstrate willful contributory infringement. Under their theory, an ISP must disconnect users when it is “substantially certain” they will continue infringing.
Cox says such a standard would make ISPs “internet police” that cut off whole households, universities, or hospitals over allegations often unverified. Cox’s lawyer Joshua Rosenkranz said the only way to protect against liability would be to cut service not only at the address where accused infringers live, but also for everyone else on the same connection. Justice Samuel Alito asked Sony’s counsel how their position would play out with regard to a university network with 70,000-user university network; Justice Sonia Sotomayor asked if it was unreasonable to cut off a five-person household based on one person engaging in repeated infringements.
The stakes go well beyond ISPs. Google, X, and other tech giants have filed briefs warning that a ruling against Cox could “wreak havoc” on the broader tech ecosystem-especially AI platforms. If courts let copyright holders sue AI providers when users generate infringing outputs, companies may have “no choice but to constrain their actions” in order to avoid liability. This is not a theoretical risk: AI systems trained on copyrighted data can indeed produce outputs that closely resemble protected works, and thus trigger infringement claims under evolving standards like the Goldsmith fair use narrowing.
Under Goldsmith, simply adding new expression or style isn’t enough; the purpose of the new work has to be meaningfully different from the original. For AI, that may mean outputs serving the same market function as the source material-say, commercial images or text-can’t pass the transformative test. A Getty lawsuit against Stability AI cited 12 million copied images and even distorted Getty watermarks, illustrating how the ingestion of training data can yield high-risk outputs. Contributory liability could expand in ways that make AI providers subject not only to suits over their own training practices but also to user prompts generating infringing content.
The Court has historically wrestled with technology inflection points like this. In the 1984 Betamax case, it shielded VCR makers from contributory infringement claims and enabled an era of home recording and time-shifting. But in MGM v. Grokster, 2005, it found file-sharing services liable when their products were marketed for infringement. The Cox case sits between these poles: ISPs provide generic connectivity, yet the plaintiffs argue Cox’s inaction amounts to facilitation.
Contributory infringement itself has developed in fits and starts in the digital environment, with patent law requiring intent to induce and some copyright decisions finding mere knowledge sufficient. Justice Neil Gorsuch suggested sending the case back to lower courts to reapply a “purpose” standard, signaling caution against expanding liability without congressional action.
From an engineering standpoint, enforcing the labels’ proposed standard would require ISPs to implement real-time infringement detection systems capable of mapping IP activity to individual users-a task complicated by Network Address Translation, dynamic IP assignment, and encrypted traffic. Content filters, watermark detection, and usage policies that AI platforms already deploy mitigate some infringement, but these mechanisms are far from foolproof. If the Court sides broadly with the labels, similar enforcement burdens could cascade across the internet economy, changing the shape of both connectivity services and generative AI operations.
