Contributory and Vicarious Copyright Liability After Cox v. Sony: Who Pays When Someone Else Infringes

In about two years, an anti-piracy vendor sent Cox Communications 163,148 notices saying that identified Cox connections were being used to share music without permission. Cox terminated 32 subscribers for infringement during the claim period. It terminated hundreds of thousands for not paying their bills. In an earlier case, its own abuse team had summarized one practice in two words and an equals sign: "DMCA = reactivate." A jury in Virginia heard the music companies' version of all this and awarded $1 billion.

On March 25, 2026, nine Justices agreed that Cox was not contributorily liable at all. Seven of them said why in a way that reaches far beyond one cable company: a business that provides a service to the public does not become a copyright infringer because it knows some customers will misuse the service and does too little to stop them.

That sentence changes the law of more than internet access. For decades, American courts had asked a copyright plaintiff to show two things against an intermediary: knowledge of the infringement and a material contribution to it. Nearly every hosting complaint, marketplace demand letter, and early AI lawsuit was built on that frame. Cox Communications, Inc. v. Sony Music Entertainment, 607 U.S. 583 (2026), removes it as a freestanding route to contributory liability. What remains is narrower and more interesting, and it rewards careful reading, because the opinion is short and several things people say about it are wrong.

The short answer

• Contributory liability now turns on intent, shown in one of two ways. The provider of a service is contributorily liable "only if it intended that the provided service be used for infringement," and that intent "can be shown only if the party induced the infringement or the provided service is tailored to that infringement."
• Knowing is not intending. Notices, even very many of them, plus continued service do not establish contributory liability. The Court rejected the Fourth Circuit's rule that supplying a product with knowledge of the recipient's infringing use is enough.
• Inducement survives intact. A provider that "actively encourages infringement through specific acts" is exposed, and early district court decisions read that phrase to reach more than advertising.
• "Tailored" means something specific. The Court defined it as not capable of substantial or commercially significant noninfringing uses, the Sony Betamax test. Courts are already disagreeing about whether to apply that test to a whole platform or to one feature.
• Vicarious liability was not decided. The Court denied Sony's petition on that theory, so the Fourth Circuit's flat-fee reasoning stands in that circuit, and the older cases on control plus direct financial benefit remain good law elsewhere. Vicarious liability does not require knowledge. For a platform that takes a cut and controls its sellers, this is now the live exposure.
• Direct infringement was not decided either. Anything the defendant itself copies, displays, or performs is outside Cox.
• Section 512 still matters, unevenly. For a pure conduit, its work has shrunk. For a host or marketplace facing direct and vicarious claims, its limits on money and injunctions remain valuable, and a provider that drops its repeat infringer program gives that protection up.
• Justice Sotomayor's route did not command a majority. Two Justices would have left open common-law aiding and abetting with a demanding, specific-knowledge version of intent. Plaintiffs will cite that opinion, but it is not the law.

The rest of this article explains how the Court got there, what it left open, and what the first seven months of lower court decisions show. For the operating steps, see the companion Assessing and Reducing Secondary Copyright Liability: A Practical Guide.

Three kinds of liability, in one table

Copyright infringement is usually discussed as if it had one defendant, the person who made the copy. In practice the person who made the copy is often a teenager with a torrent client, an anonymous overseas uploader, or a thinly capitalized shell. The party with assets is the company in the middle. Secondary liability, meaning liability for the infringement of another, is the set of doctrines that decide when that company pays.

The Copyright Act says almost nothing about it. Section 501 makes anyone who violates the exclusive rights in 17 U.S.C. § 106 an infringer (17 U.S.C. § 501(a)), and as the Court said in 1984 and repeated in 2026, the Act "does not expressly render anyone liable for infringement committed by another." The doctrines are judge-made, drawn from tort principles, and every one of them requires a direct infringement by someone.

Where the doctrine came from

Ben-Hur and the dance halls

The oldest root is Kalem Co. v. Harper Brothers, 222 U.S. 55 (1911). Kalem made a silent film of Ben-Hur without permission and sold it to dealers, knowing exhibitors would show it. Justice Holmes held Kalem liable, and his reasoning contains both halves of the modern dispute. Kalem "not only expected but invoked by advertisement the use of its films" for the infringing purpose. But, he also wrote, "mere indifferent supposition or knowledge on the part of the seller" that a buyer would act unlawfully is "not enough." A century later, Justice Thomas used those same words to decide Cox.

Vicarious liability grew up separately, in the entertainment business. Courts in the early twentieth century held dance hall proprietors liable when a hired band played unlicensed songs, even if the proprietor did not know or had told the band not to. The band brought in customers, and the proprietor controlled the premises. The contrast case was the landlord who rents a building for a fixed fee and has no say in what happens inside.

Shapiro, Bernstein & Co. v. H. L. Green Co., 316 F.2d 304 (2d Cir. 1963), turned that contrast into a rule. A department store licensed its record counter to a concessionaire, which sold bootleg records. The store did not know. But its license let it discharge concession employees in its "unreviewable discretion," and it took 10 or 12 percent of gross receipts as its compensation. A store with the power to police and a percentage of every sale, bootleg or legitimate, was responsible. Keep that percentage in mind. It is the fact that separates a commission-based marketplace from a flat-fee ISP, and it comes back below.

Gershwin Publishing Corp. v. Columbia Artists Management, Inc., 443 F.2d 1159 (2d Cir. 1971), supplied the sentence that governed contributory liability for fifty years: "one who, with knowledge of the infringing activity, induces, causes or materially contributes to the infringing conduct of another, may be held liable as a 'contributory' infringer." The defendant there was a concert manager that organized the local associations, obtained the program, printed it, and helped create the audience. Its participation was pervasive. The words "materially contributes" were written with that kind of defendant in mind, and they later traveled a long way from it.

Sony and the staple article

Then came the VCR. In Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984), studios sued the maker of the Betamax on the theory that it had supplied the means of infringement knowing that some buyers would copy programs. The Court borrowed from patent law's staple article doctrine, now in 35 U.S.C. § 271(c), which protects sellers of ordinary goods that have lawful uses. It held that selling copying equipment does not make the seller a contributory infringer if the product is "capable of substantial noninfringing uses." The Court found time-shifting, recording a show to watch later, to be such a use. It also noted that there was no precedent for liability based only on selling equipment with constructive knowledge that customers may use it to copy.

The logic of the rule matters more than the machine. A staple article rule refuses to let a copyright owner control a market in a general-purpose technology. If an article is good for nothing except infringement, nothing valuable is lost by imputing an intent to infringe. If it has real lawful uses, imputing that intent would hand rightsholders a veto over the technology.

Napster, Aimster, and the service problem

Sony involved a one-time sale. Once a Betamax left the store, Sony had no further relationship with the buyer. Internet services are different, because the provider keeps serving the user, and that difference became the key to the next generation of cases.

In A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001), the Ninth Circuit distinguished between "the architecture of the Napster system" and Napster's conduct with actual, specific knowledge of direct infringement. Napster could not hide behind Sony because it knew which files were infringing and could block them. It was vicariously exposed too: it had reserved the right to terminate users, and more infringing music drew more users. Judge Posner took a different approach in In re Aimster Copyright Litigation, 334 F.3d 643 (7th Cir. 2003), holding that encrypting users' communications so that the operator could not see what they shared was no defense, because "[w]illful blindness is knowledge, in copyright law."

Grokster

Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005), is the case everyone thinks they remember. Grokster and StreamCast distributed decentralized peer-to-peer software engineered so that neither company could see or stop what users traded. The Ninth Circuit held them protected by Sony. The Supreme Court unanimously vacated, on a new-to-copyright theory borrowed from 35 U.S.C. § 271(b): "one who distributes a device with the object of promoting its use to infringe copyright, as shown by clear expression or other affirmative steps taken to foster infringement, is liable for the resulting acts of infringement by third parties."

Three features of the opinion carry forward.

First, it listed the evidence. The defendants aimed at a known source of demand, former Napster users. They made no attempt to develop filtering tools. And their advertising revenue scaled with volume that the record showed was infringing. The Court said the third item alone would not justify an inference of intent, but viewed in context, "[t]he unlawful objective is unmistakable."

Second, it drew a limit in a footnote. "[I]n the absence of other evidence of intent, a court would be unable to find contributory infringement liability merely based on a failure to take affirmative steps to prevent infringement, if the device otherwise was capable of substantial noninfringing uses." That is footnote 12, and it decided Cox.

Third, it left a door ajar. The Court said Sony "was never meant to foreclose rules of fault-based liability derived from the common law," and it declined to revisit Sony's balance "when liability rests solely on distribution with knowledge that unlawful use will occur." Plaintiffs spent the next twenty years walking through that door. Cox closed it, over two Justices' objection.

Footnote 9 of Grokster also gave the modern description of vicarious liability, one that matters because the Court has now twice declined to revisit it: liability is allowed "when the defendant profits directly from the infringement and has a right and ability to supervise the direct infringer, even if the defendant initially lacks knowledge of the infringement."

One more detail would matter later. Justice Ginsburg, concurring, wrote that liability may be predicated on actively encouraging infringement through specific acts, or on distributing a product that distributees use to infringe if the product is "not capable of 'substantial' or 'commercially significant' noninfringing uses." In 2005 that was one concurrence's summary. In 2026 the majority cited it as the definition of both routes.

The era of knowledge plus material contribution

Between 2005 and 2026, lower courts mostly treated inducement as one theory and Gershwin as another. The second was far easier to plead.

The Ninth Circuit's version came from Perfect 10, Inc. v. Amazon.com, Inc., 508 F.3d 1146 (9th Cir. 2007): "a computer system operator can be held contributorily liable if it 'has actual knowledge that specific infringing material is available using its system,' and can 'take simple measures to prevent further damage' to copyrighted works, yet continues to provide access to infringing works" (citations omitted). The simple measures test became the standard language in demand letters to hosts and search engines. It was limited in Perfect 10, Inc. v. Visa International Service Ass'n, 494 F.3d 788 (9th Cir. 2007), where a divided panel held that credit card networks processing payments to infringing sites did not materially contribute, because payment services make infringement more profitable but add "an additional step in the causal chain." Judge Kozinski, dissenting, called payment "an essential step in the infringement process." That phrase then did its own work: Louis Vuitton Malletier, S.A. v. Akanoc Solutions, Inc., 658 F.3d 936 (9th Cir. 2011), upheld a verdict against a web host that ignored at least eighteen infringement notices and controlled what the court called the "master switch" keeping the sites online. In the other direction, Mon Cheri Bridals, LLC v. Cloudflare, Inc. (N.D. Cal. Oct. 6, 2021), granted summary judgment to a content delivery and security provider because the record did not show that its services materially contributed to the infringement.

Specific knowledge was the main limit. Luvdarts, LLC v. AT&T Mobility, LLC, 710 F.3d 1068 (9th Cir. 2013), affirmed the dismissal of a suit against wireless carriers when the plaintiffs' notices did not identify which titles were infringed, by whom, or when.

The Fourth Circuit went furthest in the Cox litigation. In BMG Rights Management (US) LLC v. Cox Communications, Inc., 881 F.3d 293 (4th Cir. 2018), it held that a "should have known" jury instruction was wrong, ordered a new trial, and reasoned that intent could be presumed where a seller knows infringement is substantially certain to result from a sale. Its example was a company that leases VCRs, learns specific customers use them to infringe, and renews the lease anyway. The Fifth Circuit reached a parallel result in UMG Recordings, Inc. v. Grande Communications Networks, LLC, 118 F.4th 697 (5th Cir. 2024), affirming the jury's liability finding (while vacating the $46,766,200 damages award for a new trial) and holding that continuing to provide internet access after notice was a material contribution. It rejected the argument that Twitter, Inc. v. Taamneh, 598 U.S. 471 (2023), changed copyright law, since Taamneh was an antiterrorism case that never mentioned copyright. The Supreme Court vacated that judgment and sent it back for reconsideration on April 6, 2026, 146 S. Ct. 2152 (2026).

How the Cox litigation worked, at the level of packets

The technology explains a good deal of the outcome.

An ISP assigns each subscriber account an Internet Protocol address, the number that identifies a connection on the network. One address may serve a household, a coffee shop, a dormitory, or, when Cox sells to a smaller regional ISP, thousands of people. The ISP can match an address to an account. It generally cannot tell which person behind the address did what, and it does not store the content moving through its network.

BitTorrent, the file-sharing protocol at issue, splits a file into small pieces. A user downloads pieces from many peers at the same time and, by default, shares pieces with others while doing so. Because participants in a torrent "swarm" can see each other's addresses, a monitoring company can join the swarm, confirm that a given address is offering a recognized file, identify the address's ISP, and send the ISP a notice. MarkMonitor did this for the record labels.

What the ISP gets, then, is a statement that someone using account X offered work Y at time Z. It has nothing to take down. Its only practical responses are to forward warnings, slow or suspend the connection, or cut off the account. The Court's opinion opens its factual account with these limits: ISPs "have limited knowledge about how their Internet services are used and who uses them," and they "cannot directly control how their Internet services are used."

Cox's response was a graduated system. On the Court's description, a warning followed the second notice for an account; further notices led to a suspension until the subscriber responded, then a suspension until the subscriber called in; and after 13 notices the subscriber was subject to termination. Cox said this ended 98 percent of identified infringement. Sony emphasized the 32 terminations and internal statements favoring subscriber revenue. The Court stated the disagreement without resolving it, and, importantly, held that it did not matter.

Cox could not rely on the DMCA safe harbor because BMG had held that it failed to reasonably implement a repeat infringer policy during the relevant period, a holding that rested heavily on the "DMCA = reactivate" emails. The claim period ran from 2013 to 2014. Sony sued for contributory and vicarious infringement of 10,017 works, and the jury found for Sony on both, found willfulness, and awarded $1 billion, about $99,830 per work.

In Sony Music Entertainment v. Cox Communications, Inc., 93 F.4th 222 (4th Cir. 2024), the Fourth Circuit split the verdict. On vicarious liability it reversed. Subscribers paid a flat fee "no matter what they did online," and Cox "would receive the same monthly fees even if all of its subscribers stopped infringing." On contributory liability it affirmed, in the sentence that became the target at the Supreme Court: "supplying a product with knowledge that the recipient will use it to infringe copyrights is exactly the sort of culpable conduct sufficient for contributory infringement." It vacated the damages and ordered a new damages trial. The Supreme Court granted Cox's petition on contributory liability and denied Sony's on vicarious liability.

What Cox actually held

The opening move: reluctance

Justice Thomas, joined by the Chief Justice and Justices Alito, Kagan, Gorsuch, Kavanaugh, and Barrett, began with the statutory silence. Congress ordinarily imposes secondary liability expressly, citing Central Bank of Denver, N.A. v. First Interstate Bank of Denver, N.A., 511 U.S. 164 (1994), the case that refused to imply aiding and abetting liability under the securities laws. The Court's precedents recognize specific forms of secondary copyright liability "that predated the Copyright Act," but "we are loath to expand such liability beyond those precedents."

Read that against the structure of the opinion. The Court treats secondary copyright liability as a closed set of historical categories. New categories need Congress.

The rule: intent, shown two ways

Then come the two sentences that matter most:

"The provider of a service is contributorily liable for the user's infringement only if it intended that the provided service be used for infringement. The intent required for contributory liability can be shown only if the party induced the infringement or the provided service is tailored to that infringement."

Each "only" does work. The first makes intent the element. The second closes the list of ways to prove it. Lower courts have noticed: as Judge Stein put it in the OpenAI multidistrict litigation, the Court "held that 'material contribution' is not a valid theory of contributory liability for copyright infringement," in In re OpenAI, Inc. Copyright Infringement Litigation (Ziff Davis, Inc. v. OpenAI, Inc.) (S.D.N.Y. Aug. 6, 2026).

Inducement. "A provider induces infringement if it actively encourages infringement through specific acts." The Court's examples are Grokster, where the companies "promoted and marketed their software as a tool to infringe copyrights," Kalem, and Henry v. A. B. Dick Co., 224 U.S. 1 (1912), the last with the note that it was overruled on other grounds. Those are examples, a distinction that became important within weeks.

Tailoring. "A service is tailored to infringement if it is 'not capable of "substantial" or "commercially significant" noninfringing uses,'" quoting the Ginsburg concurrence and Sony. The Court then retold Sony: the tape recorder could record shows for later viewing, and it could also be used to reproduce and sell copyrighted programming, and that first capability was enough.

The Court tied both to patent law, with which copyright has what Sony called a "historic kinship": inducement to 35 U.S.C. § 271(b), and tailoring to § 271(c), which reaches someone who sells a component "knowing the same to be especially made or especially adapted for use in an infringement." That analogy is a gift to litigators on both sides, because patent law has a large body of cases on both provisions. MC Law's Indirect and Divided Patent Infringement: Inducement, Contributory Liability, and the Akamai Problem describes that law, including the knowledge requirement of Global-Tech Appliances, Inc. v. SEB S.A., 563 U.S. 754 (2011), under which willful blindness requires a subjective belief in a high probability of the fact and deliberate action to avoid learning it. How far the patent cases transfer is an open question, and the first courts to try it have not said.

Three cases against mere knowledge

The Court then quoted three decisions to establish that knowledge is not intent: Kalem's "mere indifferent supposition or knowledge"; Sony's statement that there is no precedent for liability based only on selling equipment "with constructive knowledge"; and Grokster's footnote 12. Notice what the quotation of footnote 12 leaves out. The original footnote began "in the absence of other evidence of intent" and ended "if the device otherwise was capable of substantial noninfringing uses." The Fourth Circuit had seized on the first qualifier to say that knowing supply was the other evidence. The majority's answer is that it is not. But "other evidence of intent" remains a real category. Failure to filter can still corroborate inducement in the way Grokster described; it just cannot substitute for it.

The application

The application takes a page. Cox did not induce: Sony offered "no 'evidence of express promotion, marketing, and intent to promote' infringement," and Cox "repeatedly discouraged copyright infringement by sending warnings, suspending services, and terminating accounts." Cox did not tailor: its service was "clearly" capable of substantial noninfringing uses. "Cox did not tailor its service to make copyright infringement easier. Cox simply provided Internet access, which is used for many purposes other than copyright infringement."

The Fourth Circuit's holding "went beyond the two forms of liability recognized in Grokster and Sony" and conflicted with the "repeated admonition that contributory liability cannot rest only on a provider's knowledge of infringement and insufficient action to prevent it."

The DMCA answer

Sony's best structural argument was that the safe harbor would be pointless if ISPs were never liable for serving known infringers, so Congress must have assumed they could be. The Court's response ran five sentences: "Sony overreads the DMCA." The statute does not expressly impose liability. It "merely creates new defenses." And 17 U.S.C. § 512(l) says that failing to qualify for a safe harbor "shall not bear adversely upon" a defense that the conduct is not infringing.

The bad facts that did not matter

It is easy to overlook how much bad evidence the Court recited and passed over. The Fourth Circuit had called the record sufficient to show that Cox knew of repeat infringement, traced it to specific users, and kept them on for revenue, with emails a jury could read as contempt for the law. None of that fit a theory of encouragement. Internal contempt for the notice process, on this reading, is not active encouragement of infringement. That is a remarkable result, and it should not be overread: the same internal statements attached to a product decision, a sales pitch, or a marketing plan would look very different.

Justice Sotomayor's concurrence

Justice Sotomayor, joined by Justice Jackson, concurred only in the judgment. Her opinion deserves careful reading because it is both a roadmap of the majority's limits and a warning about its consequences.

Her objection to the "only." She agrees with the first sentence of the rule and says the majority "errs in the very next sentence." Sony and Grokster preserved other common-law forms of liability, and Grokster said so expressly. Stare decisis, she adds, citing Kimble v. Marvel Entertainment, LLC, 576 U.S. 446 (2015), requires applying what those cases said, and the Central Bank principle is beside the point because the Court already held in 1984 that the Copyright Act impliedly provides for secondary liability.

Her DMCA objection. She argues that the majority "completely upends" the balance Congress struck and "consigns the safe harbor provision to obsolescence." She gives two hypotheticals: an ISP that sells a connection to a company it knows runs a website that exclusively hosts pirated works, and one that sells to a customer who walks in and explains that the more scrupulous ISP in town cut him off for piracy. Under the majority's rule, she says, neither faces liability. She notes that Cox's counsel agreed at argument that the safe harbor would not "d[o] anything at all" going forward.

Her alternative. She would decide the case under common-law aiding and abetting, relying on Taamneh and Smith & Wesson Brands, Inc. v. Estados Unidos Mexicanos, 605 U.S. 280 (2025). That route requires conscious, culpable participation, but it recognizes that intent can be inferred when an actor knows that a consequence is certain or substantially certain, drawing on the Restatement (Second) of Torts § 8A. She gives the example of handing a gun to someone you know will shoot.

Why Cox still wins on her view. Notices identify a connection, not a person. "That informational gap is fatal here." After three notices to a family home, Cox knows only that someone may infringe again; it does not know who, and the infringer could be a neighbor with the Wi-Fi password. The gap is wider for regional ISPs, universities, hospitals, and military bases, which she compares to the middlemen in Smith & Wesson. On this record plaintiffs showed at most indifference, and indifference is not enough.

Her footnote. Footnote * contains perhaps the most useful sentence for anyone reading the case in practice: "Under the majority's view, ISPs could still face secondary liability if they design their service to facilitate infringement or if they promote their service as infringement friendly." That is one Justice's description of a rule she opposes, so it is not a holding. It is, though, a fair statement of what the majority's two routes would catch, and plaintiffs will quote it.

What Cox did not decide

The opinion is about ten pages long, and it leaves real gaps. These are the ones that will generate the next round of litigation.

Is the rule limited to services offered "to the general public"?

The Court's introductory statement of its holding is that a company is not liable "for merely providing a service to the general public with knowledge that it will be used by some to infringe." Later statements of the rule drop the qualifier. Does it matter? Many secondary liability defendants are not providing general services to the public. A company might hand photographs to one retailer, hire one developer, or sell custom services to one known infringer. The rule's first sentence speaks of "[t]he provider of a service." Whether it governs a defendant that is not a provider of a service at all, such as an individual executive or a licensor, is not addressed. One court has already applied Cox to a fashion wholesaler's dealings with its retailers (Ravazzani v. C K Collection, Inc. (S.D.N.Y. Sept. 28, 2026)), which suggests the general rule will spread beyond service providers. But a defendant in a one-to-one relationship is a much poorer fit for the staple article logic, and the older cases on hiring and personal participation have not been overruled in terms. Erickson Productions, Inc. v. Kast, 921 F.3d 822 (9th Cir. 2019), for example, upheld a contributory verdict on plain-error review against a business owner whose developer used unlicensed photos on his site. Its instruction allowed liability on "reason to know," and that part of its reasoning is now hard to square with Cox.

Does knowledge remain a separate element?

The Court spoke of intent and said nothing about whether knowledge of specific infringement remains a requirement alongside it. The first courts differ in how they state the test. The Southern District of Florida describes two elements, direct infringement and intentional inducement or encouragement (United King Film Distribution (1990) Ltd. v. MoonPay USA LLC (S.D. Fla. July 13, 2026)). The Middle District of Pennsylvania noted that Cox "appeared to describe contributory copyright infringement as a two-element claim" but kept the Third Circuit's knowledge element in its analysis (Amazon Content Services LLC v. Weibley (M.D. Pa. June 10, 2026)). The Northern District of California kept knowledge of specific infringement too (Nazemian v. NVIDIA Corp., No. 24-cv-01454-JST, ECF No. 303 (N.D. Cal. May 5, 2026)). The practical stakes are smaller than they look, because a plaintiff who can prove inducement will usually show awareness of infringement along the way, and Grokster itself did not demand knowledge of each specific act. Defendants should still preserve the point.

What is the unit of analysis for "tailored"?

This is the most consequential disagreement so far. In MoonPay, plaintiffs argued that a cryptocurrency widget "as deployed through Israel TV's payment flow" was especially adapted for infringement. The court said the argument "ignores the standard set forth in Cox, which asks only whether the service is capable of noninfringing use." Plaintiffs had not disputed that the overall platform had lawful uses. In Nazemian, NVIDIA argued that its NeMo Megatron framework had many lawful uses. The court accepted plaintiffs' framing and looked only at the alleged scripts for downloading and preprocessing a dataset containing pirated books, which were alleged to have "no other purpose than to speed up the process of infringement." Same test, opposite unit of analysis.

The two decisions can be reconciled. MoonPay was on summary judgment, the widget was a general payment tool that one bad actor misused, and the plaintiffs conceded general capability. Nazemian was on the pleadings, and the scripts were written by the defendant and directed at a particular dataset. But the question will not stay this tidy. Every platform has components. A court that slices narrowly enough can find a component with no lawful use, and a court that refuses to slice gives immunity to any infringing feature bundled with a lawful one. Patent law has been here before: § 271(c) asks about the "component," not the larger machine. Plaintiffs should plead the narrowest separable thing the defendant built and offered. Defendants should argue that a separable function must actually be separately offered.

Does "tailored" have any design content, or is it only about capability?

The Court's definition is capability, and MoonPay took it literally: "The inquiry turns on capability, not actual use." Yet the opinion also says Cox "did not tailor its service to make copyright infringement easier," which sounds like a question about design. Justice Sotomayor reads the majority to reach services designed to facilitate infringement. Suppose a general-purpose file locker adds a one-click "re-upload to a new link when a takedown arrives" feature. The locker can obviously store lawful files, so under a capability reading there is no tailoring. The feature, though, has an obvious and rather specific function. A plaintiff would plead the feature as the tailored service, or as an inducing specific act, and that is probably the better fit. Designing a tool specifically to promote infringement is the very thing Nazemian treated as inducement.

Must inducement involve a message?

Cox describes Grokster in terms of promotion and marketing, which led NVIDIA to argue that a complaint without advertising fails. Judge Tigar disagreed: advertising or promotion "is an example of an inducing act, not a pre-requisite." Grokster supports him. It said proving a message was sent is "the preeminent but not exclusive way" of showing active steps, and it named instructing users how to infringe as an example. Weibley went further, treating the operation of a cheap pirate streaming service, plus concealment and domain-hopping, as encouragement through specific acts. That was a default judgment against an operator who was plainly a direct or near-direct actor, so it should not be stretched.

The limits have started to appear too. In Ravazzani, a brand's Instagram account "liked" a third party's infringing post. The court held that a reasonable jury could not conclude from the like alone that the brand intended to induce infringement, particularly since the like came after the post and nobody knew who clicked it. In MoonPay, refusing to keep disabling API keys after initially doing so was not encouragement: the "pipeline" argument was "foreclosed" by the Court's admonition about insufficient action.

What about willful blindness?

Cox's holding removes the point of much willful blindness doctrine in contributory cases: if actual knowledge is not enough, its legal equivalent is not enough either. Blindness can still matter as evidence. Engineering a system so that you cannot see infringement was part of the Aimster story and part of the Grokster record, and it may be an affirmative step that shows purpose. It also still matters under § 512(c) and to willfulness under 17 U.S.C. § 504(c)(2).

Trademark and patent are different

Cox is a copyright decision. Contributory trademark liability rests on Inwood Laboratories, Inc. v. Ives Laboratories, Inc., 456 U.S. 844 (1982), which expressly covers continuing to supply a product to someone one knows or has reason to know is infringing. The Ninth Circuit applied that test to an online marketplace in Y.Y.G.M. SA v. Redbubble, Inc., 75 F.4th 995 (9th Cir. 2023), requiring knowledge of specific infringers or instances. A marketplace that treats Cox as a universal shield will be surprised by a trademark complaint. See Trademark Liability Online: Keyword Advertising, Marketplaces, and Contributory Infringement After Tiffany v. eBay. Whether Inwood's rule survives the same logic is a question defendants will start asking, but no court has so held in anything we found.

What lower courts have done so far

Cox is about seven months old as of this writing, and the decisions are mostly from district courts. These are the ones reviewed for this article.

Three things stand out.

Old formulas die slowly. Broadcast Music, Inc. v. QM Entertainment, LLC (W.D. Wash. Apr. 10, 2026), a default judgment against a bar over live music, recited "induces, causes or materially contributes" two weeks after Cox without citing it. Default judgments rarely test anything, but the example shows why parties need to cite the decision to the court themselves.

Plaintiffs are retreating from the theory, sometimes visibly. Beyond Ziff Davis, trade press and docket reports describe voluntary withdrawals of contributory claims by the New York Times and other news plaintiffs against OpenAI, and by music publishers against Anthropic. Labels reportedly dismissed their parallel suits against Verizon and Altice. In Grande, the post-remand letters described in press reports were about costs and disposition, with the labels accepting that Cox changed the governing law. These accounts come from filings and press reports, not from rulings reviewed here, so check the dockets before relying on them.

AI complaints are not uniformly losing. Press reports state that in Disney Enterprises, Inc. v. MiniMax, No. 2:25-cv-08768 (C.D. Cal.), Judge Blumenfeld declined in May 2026 to dismiss direct and secondary claims against the operator of the Hailuo video generator. We reviewed the defendant's reply brief but could not obtain the order itself, so this article does not report how the order treated Cox.

Vicarious liability, the theory that did not move

Vicarious liability has two elements: the right and ability to supervise the infringing conduct, and a direct financial interest in it. Neither involves knowledge or intent. It is the closest thing in copyright law to strict liability for someone else's act, which is why, after Cox, plaintiffs have every reason to plead it and defendants have every reason to worry about it.

The Court said nothing about the merits, having denied Sony's petition. Defendants will still argue that Cox's reluctance to expand secondary liability bears on how far courts extend these older cases. The argument has some force at the edges. It has none against the core, since the opinion recognizes vicarious liability as one of the "two categories" and dance hall liability is about as historic as secondary liability gets.

The financial interest element

The Fourth Circuit's opinion is now the most careful treatment of the element for flat-fee services. Its core sentence: "the crux of the financial benefit inquiry is whether a causal relationship exists between the infringing activity and a financial benefit to the defendant." Monthly internet fees did not qualify, even though Cox weighed subscriber revenue when deciding whether to terminate ("he pays 317.63 a month"). Peer-to-peer traffic volumes did not show that anyone subscribed in order to infringe. Higher payments by infringing subscribers did not either, because expensive tiers have many lawful explanations. This remains law in the Fourth Circuit.

The Ninth Circuit asks whether infringing material "acts as a draw." Fonovisa, Inc. v. Cherry Auction, Inc., 76 F.3d 259 (9th Cir. 1996), held that it did for a swap meet that collected admission and parking fees from customers coming for bargain counterfeit tapes. Ellison v. Robertson, 357 F.3d 1072 (9th Cir. 2004), held that a benefit need not be large but must be causally connected, and found no evidence that AOL gained or lost subscribers because of infringing Usenet posts. Perfect 10, Inc. v. Giganews, Inc., 847 F.3d 657 (9th Cir. 2017), added an important limit: the plaintiff must connect the benefit to its own works, not to infringement in general. Erickson held that merely saving a licensing fee is not a direct financial benefit. The Second Circuit takes a similar view of draw evidence in EMI Christian Music Group, Inc. v. MP3tunes, LLC, 844 F.3d 79 (2d Cir. 2016).

Now return to the H.L. Green percentage. Courts and the Senate report treat flat periodic payments as ordinarily not a direct benefit. A service that takes a commission on the very transaction in which the infringing item is sold has a different structure. The infringing sale itself generates the money. A flat-fee host can say it would earn the same if everyone stopped infringing. A marketplace taking a percentage cannot say that about the disputed listing.

The control element

In the Ninth Circuit, control requires both "a legal right to stop or limit the directly infringing conduct, as well as the practical ability to do so" (Amazon.com). Google lacked that over third-party sites. Visa could cut off payments but could not stop the sites from infringing, and pressure is not control. Napster had control, within "the boundaries of the premises that Napster 'controls and patrols,'" meaning its own index. Luvdarts held that a carrier's theoretical ability to build a supervisory system does not count; the test looks to the system as it currently is. Concord Music Group, Inc. v. X Corp. (M.D. Tenn. Mar. 5, 2024) rejected vicarious liability against a social platform for lack of control over its users, even as it let part of the contributory claim go forward. Nazemian dismissed the vicarious claim because control over one's own tools is not control over what customers independently do.

A contract right to terminate helps plaintiffs but does not end the inquiry, which is why the sentence in the Cox background about ISPs lacking direct control is worth quoting if you represent a conduit. For a host or marketplace, the facts are less friendly. It stores the file, can delete it, and writes the rules. In the swap meet and department store cases, that contractual and practical authority was enough.

Inducement: what a specific act looks like now

Since inducement is now the main contributory route, everyone's documents become relevant. Grokster's evidence, adapted to current practice, falls into a few groups.

Messages. Advertising an infringing use, tutorials showing how to obtain protected works, demo reels built on recognizable characters, customer support answers telling users how to play the pirated movie. In Grokster, responding to users' requests for help was part of the record. In Columbia Pictures Industries, Inc. v. Fung, 710 F.3d 1020 (9th Cir. 2013), the Ninth Circuit extended inducement beyond "devices" to services, set out four elements (distribution of a device or product, acts of infringement, an object of promoting its use to infringe, and causation), and affirmed liability against a torrent site operator on an extensive record of encouragement.

Targeting. Aiming at a known source of infringing demand, the way Grokster and StreamCast aimed at Napster's users.

Design. Features whose evident point is to help infringement along, such as NVIDIA's alleged scripts. Filtering failures belong here only as corroboration after Cox.

Business model. Advertising revenue that scales with infringing volume. Grokster said this alone would not suffice, and that remains true.

Selling leniency. One pre-Cox decision deserves new attention. In Concord v. X Corp., the court allowed claims to proceed on three alleged practices: letting users pay for more forgiving treatment under the infringement policy, unreasonably delaying takedowns to make the platform more attractive to infringers, and failing to act against severe serial infringers. After Cox, the second and third look like "insufficient action." The first is different: selling a paid tier that buys tolerance of infringement is an affirmative commercial act that speaks directly to the user. If that fact pattern is proved, it is among the better inducement theories available. (The opinion predates Cox and relied on BMG, so check how that case has developed before relying on it.)

Employees. Capitol Records, LLC v. Vimeo, Inc. (2d Cir. Sept. 9, 2025), an order on rehearing, recognizes that encouraging users to make infringing "lip-dub" videos is a separate, Grokster-based way of arguing that a provider has the right and ability to control under § 512(c)(1)(B), though plaintiffs had waived it there. Staff encouragement can matter both for liability and for the safe harbor.

Hypothetical: the fiber salesman. Suppose Northloop Fiber, a small regional ISP, runs an ordinary service with an ordinary acceptable use policy. Under Cox, the notices it receives do not make it contributorily liable, whatever it does with them. Now suppose its sales team, chasing students near a university, hands out a flyer saying "No throttling, no letters from the labels, download whatever you want." The company's lawyer would want that flyer gone yesterday. The service is still capable of lawful uses, so tailoring is unavailable. But the flyer is a specific act, it is aimed at a demand for infringement, and it converts a thousand previously harmless notices into evidence of who responded to the message. Whether a single flyer proves a corporate object is a fair question, and Grokster involved far more. But Northloop has moved from a claim that should be dismissed on the pleadings to one that may well get to discovery. It also now has a problem under Grokster's causation requirement, in that the flyer's audience is easy to identify.

Does section 512 still matter?

Justice Sotomayor said Cox makes the safe harbor obsolete. For a pure conduit that may be close to true. For most other online businesses it is wrong, for reasons worth setting out.

Section 512 limits liability on every theory, not only contributory. 17 U.S.C. § 512(c)(1) says a qualifying provider "shall not be liable for monetary relief" for infringement "by reason of the storage at the direction of a user." A host still faces direct infringement claims and vicarious claims, as in McGucken v. Shutterstock, Inc., No. 23-7652 (2d Cir. Feb. 10, 2026), where a photographer pleaded direct, contributory, and vicarious theories together. Those are the claims Cox leaves alive, and § 512 can cover them.

Section 512(j) confines injunctions. A provider entitled to the safe harbor can be enjoined only in the limited forms that subsection lists, such as restraining access to infringing material at a particular online site or terminating specified accounts, and the court must weigh burden, technical feasibility, and less burdensome alternatives. A conduit faces narrower forms still. A defendant outside the harbor faces the general power of 17 U.S.C. § 502.

Red-flag and willful blindness rules give a host predictability. Viacom International, Inc. v. YouTube, Inc., 676 F.3d 19 (2d Cir. 2012), tied disqualifying knowledge to specific infringing items and explained that the "right and ability to control" requires "something more" than the ability to remove material. UMG Recordings, Inc. v. Shelter Capital Partners LLC, 718 F.3d 1006 (9th Cir. 2013), and Capitol Records, LLC v. Vimeo, Inc., 125 F.4th 409 (2d Cir. 2025), reinforce both points. Most recently, Athos Overseas Ltd. v. YouTube, Inc., No. 23-13156 (11th Cir. Jan. 7, 2026), held that YouTube's hash-matching tools, which rightsholders operate and which cannot judge legality, did not give it actual or red-flag knowledge, and that its content moderation features did not amount to the right and ability to control. Those decisions protect a host when Cox does not.

The safe harbor is not automatic, and the repeat infringer condition has teeth. Every safe harbor requires a provider to have "adopted and reasonably implemented" a policy for terminating repeat infringers (§ 512(i)(1)(A)). BMG's holding that "repeat infringers" are not limited to adjudicated infringers was untouched by Cox. The standard is forgiving: in McGucken, the Second Circuit adopted the Ninth Circuit's test from Perfect 10, Inc. v. CCBill LLC, 488 F.3d 1102 (9th Cir. 2007), and held that failing to terminate one repeat infringer among millions of contributors is not unreasonable, and Ventura Content, Ltd. v. Motherless, Inc., 885 F.3d 597 (9th Cir. 2018), held that a site run by its owner and one contractor could rely on termination criteria that were never written down. But a provider that dismantles the program because of Cox loses that protection altogether. McGucken also vacated summary judgment because fact questions remained on whether the infringement occurred "by reason of the storage at the direction of a user" and whether the marketplace had the right and ability to control. For the compliance details, see Digital Millennium Copyright Act Safe Harbors for Online Service Providers and DMCA Safe Harbor Compliance Checklist: A Practical Checklist.

Anti-infringement conduct is now evidence on the merits. The Court's first reason for finding no inducement was the absence of promotion. Its next was that Cox "repeatedly discouraged" infringement. MoonPay did the same with terms of use and domain blocking. A program that sends warnings and terminates accounts therefore helps even a provider that does not need the harbor. Abandoning it is legally available to a pure conduit, as Justice Sotomayor said, but it would give a plaintiff a tidier story in any later inducement dispute.

Inducement and the safe harbor can coexist, in theory. Fung rejected the idea that inducement automatically disqualifies a defendant. It said the court was "not clairvoyant enough to be sure" that a liable inducer could never qualify, while noting that the behavior will usually defeat a condition, such as red-flag knowledge or the right and ability to control. Justice Sotomayor's footnote says the safe harbor "is not intended to help" an ISP on either theory.

Section 512 also contains things that are not defenses. Takedown notices remain the normal way to remove material from a host. Misrepresentation liability under § 512(f) and Lenz v. Universal Music Corp., 815 F.3d 1145 (9th Cir. 2016), still constrains notice senders. And the subpoena power of § 512(h) continues, but only where a takedown notice could be sent: In re Subpoena of Internet Subscribers of Cox Communications, LLC (9th Cir. Aug. 15, 2025), joined RIAA v. Verizon Internet Services, Inc., 351 F.3d 1229 (D.C. Cir. 2003), in holding that a conduit cannot be subpoenaed that way because there is nothing for it to take down. Together with Cox, that leaves rightsholders with a Doe complaint and ordinary discovery if they want a subscriber's identity. MC Law's Identifying an Anonymous Defendant Online: A Practical Guide covers that route, including the Cable Act's consent and court-order provisions in 47 U.S.C. § 551.

Direct infringement: the theory waiting offstage

If secondary liability narrows, plaintiffs will try harder to characterize the intermediary as the copier. The limit on that is the volitional conduct requirement, which Giganews explains is really a causation requirement. It began with Religious Technology Center v. Netcom On-Line Communication Services, Inc., 907 F. Supp. 1361 (N.D. Cal. 1995). CoStar Group, Inc. v. LoopNet, Inc., 373 F.3d 544 (4th Cir. 2004), treated an ISP passively storing user-posted photos as a conduit. Cartoon Network LP v. CSC Holdings, Inc., 536 F.3d 121 (2d Cir. 2008), looked to "the volitional conduct that causes the copy to be made." VHT, Inc. v. Zillow Group, Inc., 918 F.3d 723 (9th Cir. 2019), and BWP Media USA, Inc. v. T&S Software Associates, Inc., 852 F.3d 436 (5th Cir. 2017), applied the requirement to platforms. American Broadcasting Cos. v. Aereo, Inc., 573 U.S. 431 (2014), cuts the other way when a service looks like a cable system. And Emmerich has just held, in the Fifth Circuit, that a news aggregator that frames content from the owner's server does not "transmit" a display, resting on statutory text rather than the Ninth Circuit's server test.

There is an irony here. Courts developed volition partly on the assurance that secondary liability would catch bad actors. Emmerich itself repeats that point, citing Cox. A defendant who wins on both volition and Cox may be left with no copyright exposure for conduct that a court would once have called contributory. Whether courts adjust either doctrine in response is something to watch, and some plaintiffs already argue for it.

State of play by sector

Internet access providers

For a conduit that simply sells access, contributory liability based on notices is over. Vicarious liability looks weak under the Fourth Circuit's flat-fee reasoning and the ISP's lack of control. Grande went back to the Fifth Circuit. That leaves only inducement by marketing or design, which is why the salesperson matters more than the abuse desk. Business customers remain subject to whatever contract and upstream obligations they accepted.

Hosts, CDNs, and cloud infrastructure

Akanoc's "master switch" reasoning is hard to reconcile with Cox as a basis for copyright liability, since continuing to host after notice is the same knowledge and insufficient action the Court rejected. (Its trademark holding rests on Inwood and is a separate matter.) Justice Sotomayor's example of an ISP serving a known piracy site suggests that even serving an obvious pirate is not enough under the majority's rule. There are three cautions. A host stores the file, so direct infringement allegations are possible, and § 512(c) matters. A host that partners with, promotes, or specially configures service for a pirate customer has moved toward inducement. And injunctions can reach hosts in other ways, as discussed next.

The injunction path around Cox

Fed. R. Civ. P. 65(d)(2) binds parties and those "in active concert or participation" with them who receive actual notice. Rightsholders win default judgments against pirate operators and serve them on intermediaries. Weibley approved an injunction extending to registrars and hosting providers of the infringing domains. MoonPay recounts the limit: a New York court denied contempt against MoonPay because the pirate had no account with it and MoonPay did not know which wallets were the pirate's. Expect this route, which needs no finding of secondary liability against the intermediary, to be used more often. Providers served with such orders should get advice quickly, and the scope of the "active concert" language is the fight to have.

Marketplaces and print-on-demand services

These businesses may have the most residual risk. They usually hold the listing, take a commission, set the rules, and sometimes print and ship the goods. The commission is the H.L. Green fact, and the rules plus delete button supply an argument on control. Whether the platform is a direct seller is its own question; Milo & Gabby LLC v. Amazon.com, Inc., 693 F. App'x 879 (Fed. Cir. 2017), held Amazon was not the seller of third-party goods it fulfilled, in a nonprecedential decision. A service that actually manufactures an infringing item is in a far weaker position than one that merely lists it. McGucken shows that licensing as principal does not disqualify a provider from being a "service provider," but it can raise fact issues under the storage and control requirements.

App stores

We found no reported decision applying Cox to an app store, so this is analysis, not reporting. A general store is plainly capable of lawful use, and tailoring is out unless a specific store feature is at issue. Inducement requires something more than listing an app. Featuring a stream-ripper in an editorial collection, or responding to developer questions about how to get around rights restrictions, would be different. The vicarious theory has more traction, because store operators generally have strong contractual control and collect a percentage of paid transactions. Expect the argument to be about whether the commission flows from the infringement or the app's general utility, and whether removing an app amounts to control over what the app's users do. An app that bakes in infringing content presents a stronger claim than a tool that users may misuse.

Payment companies

Visa now looks prescient. MoonPay adds a crypto on-ramp to the list, on facts that include terms of use forbidding infringing transactions, no post-transaction monitoring, and undisputed lack of contact with the pirate. It also shows that the state-law workarounds did not succeed there: Florida deceptive practices and negligence claims escaped preemption under 17 U.S.C. § 301, then failed on the merits because terms that govern users' conduct are not a promise to police, private policies cannot deceive consumers who never see them, and foreseeability does not create a duty.

AI developers

Generative AI secondary liability divides into two questions. For model training, the developer is usually accused of direct copying, and Cox has little to say. MC Law's Copyright Infringement Claims Against Generative AI: The New York Times, Getty, and What Comes Next covers that territory. For outputs, the question is who made the copy when a user prompted a model to produce a protected character or lyric. If the user is the direct infringer, the developer needs inducement, tailoring, or vicarious liability to be reached.

Tailoring will rarely work against a general model, which is obviously capable of noninfringing uses, although Nazemian shows that a dataset-specific tool is a different target. A model's failure to filter is, after Cox, not enough. That leaves marketing and instruction. The briefing in Disney v. MiniMax shows how this will be argued: the plaintiffs point to marketing lines and social posts featuring recognizable characters, and the defendant replies that a slogan is not an instruction, that 13 posts out of thousands are not a campaign, and that nobody who infringed is shown to have seen them. The causation point deserves attention, since Grokster speaks of "resulting acts of infringement."

Hypothetical: the demo reel. Suppose Quillon Sound, a startup offering a music generator, trains on licensed catalogs and blocks prompts naming living artists. Six months later, a growth manager, seeing that blocked prompts are a top complaint, publishes a "prompt tips" post explaining how to describe a famous singer's vocal style by era and label to get past the filter, with audio examples that sound like that singer's biggest hit. The company's own filter shows it can tell what users are doing. The post tells them how to get around it. Whether those outputs are infringing is a separate and difficult question, since style alone is not protected. But assuming some are, Quillon has gone from a defendant whose only sin is imperfect filtering, which Cox protects, to a defendant with a written instruction aimed at a specific kind of output. That is the difference between a claim dismissed and a claim that reaches internal messages.

What rightsholders can still do

Rightsholders have lost a theory but not the field.

Sue the infringer and take the injunction. Operators of pirate services remain directly liable, as do many uploaders, and Weibley shows how an injunction can reach the infrastructure. Cox itself notes that suits against individuals remain available, citing a file-sharing case, though it acknowledges how little they accomplish at scale. Identification is harder than it sounds: Cobbler Nevada, LLC v. Gonzales, 901 F.3d 1142 (9th Cir. 2018), held that a bare allegation that a person is the registered subscriber of an IP address does not state a claim, whether direct or contributory.

Plead inducement with facts. Identify the specific acts, the people who made them, the audience, and the connection to infringement of your works. Rule 8 and Rule 11 apply. Discovery is the prize, since internal communications are often where intent lives, and Grokster treated them as such. Conclusory allegations that a defendant "induced" will not get you there.

Plead tailoring narrowly and only if you mean it. Identify the component or function, why it has no commercially significant lawful use, and that the defendant offered it as such.

Plead vicarious liability with the two elements in mind. Conclusory draw allegations failed in Nazemian, and Giganews requires a link to your works. Evidence that customers came for your content, that they left when it was removed, or that a commission flowed from the sale itself is what matters.

Evaluate direct liability honestly. The defendant's own copying, curation, selection, and display are the strongest facts available, and they are outside Cox.

Do not stop sending notices. A compliant notice still removes content from hosts and still starts the clock on a § 512(c) provider's expeditious response. It remains useful evidence of knowledge for the safe harbor, even if it no longer establishes liability. See How to File a DMCA Takedown Notice (And Respond to One).

Mind the basics. Registration is a precondition to suit for United States works (17 U.S.C. § 411(a); Fourth Estate Public Benefit Corp. v. Wall-Street.com, LLC, 586 U.S. 296 (2019)). Statutory damages and fees depend on timely registration under 17 U.S.C. § 412, and Ravazzani flagged an unresolved question about § 412 when the contribution predates registration and the direct infringement follows it. MoonPay is a reminder that thin proof of ownership invites a fight before anyone reaches intent. The limitations period is three years (17 U.S.C. § 507(b)), with the discovery-rule question left open in Warner Chappell Music, Inc. v. Nealy, 601 U.S. 366 (2024). Statutory damages run per work, not per act of infringement, under 17 U.S.C. § 504(c)(1), with all parts of a compilation or derivative work counted as one, an issue the Fourth Circuit discussed in Sony without choosing among the competing tests. The jury decides the amount under Feltner v. Columbia Pictures Television, Inc., 523 U.S. 340 (1998). For the larger picture, see Copyright Damages and Remedies: Statutory Damages, Profits, Injunctions, and Fees.

Ask Congress. The opinion's message is that Congress expands secondary liability if anyone does. The Copyright Office's 2020 report on section 512 had already argued that the system was out of balance for rightsholders. As of this writing we are not aware of enacted legislation responding to Cox, and any proposal would need to confront the same concerns about general-purpose technology that drove the decision.

Three misreadings to avoid

"Cox means platforms never have to respond to notices." It means notice plus inaction does not establish contributory liability. A host that ignores notices loses the safe harbor and then defends direct and vicarious claims on the merits, with the ignored notices in evidence.

"Cox only protects ISPs." The rule is stated for the provider of a service, and courts have applied it to crypto on-ramps, AI developers, and even a clothing wholesaler.

"Sotomayor's substantially-certain test is still available." Seven Justices rejected any third route. Her analysis is useful for what it says about intent and specificity, and if a court someday revisits the "only," it may matter. It is not a basis for a claim today, though a plaintiff might reasonably plead facts satisfying it to preserve the argument.

Further reading

Useful secondary sources include Melville B. Nimmer and David Nimmer, Nimmer on Copyright, § 12.04 (the section the Grokster Court cited, in its current edition); William F. Patry, Patry on Copyright; the U.S. Copyright Office's Section 512 of Title 17: A Report of the Register of Copyrights (May 2020); Douglas Lichtman and William Landes, Indirect Liability for Copyright Infringement: An Economic Perspective, 16 Harv. J.L. & Tech. 395 (2003); and the Congressional Research Service's Legal Sidebar on Cox, LSB11416. For opinions cited here, UpLaw has the text and citator.

This article is general information as of 3 November 2026 and is not legal advice for any specific situation. Readers with a live exposure question or a rights enforcement decision may want to talk it through with counsel who can look at their product, their documents, and their circuit's law.

Related Documents

Related Articles

• Section 230 and Platform Liability: What the Statute Actually Says and Where It Stops: why section 230 does not cover federal copyright claims.
• Section 230 Reform and Platform Liability for User-Generated IP Infringement: where copyright and section 230 overlap in reform debates.
• Copyright Infringement Claims Against Generative AI: The New York Times, Getty, and What Comes Next: the direct infringement side of the AI cases.
• Copyright Damages and Remedies: Statutory Damages, Profits, Injunctions, and Fees: what the money looks like if liability attaches.
• Trademark Liability Online: Keyword Advertising, Marketplaces, and Contributory Infringement After Tiffany v. eBay: the parallel trademark doctrine that Cox did not change.
• Indirect and Divided Patent Infringement: Inducement, Contributory Liability, and the Akamai Problem: the patent provisions the Court used as its model.
• Federal Subject-Matter Jurisdiction: Arising Under, Diversity, Supplemental Claims, and the Defect That Never Waives: why copyright claims belong in federal court and how state-law companion claims fit (same batch).
• The Computer Fraud and Abuse Act After Van Buren: Gates, Insiders, Scrapers, and the Loss Requirement: another Supreme Court decision narrowing a broad liability statute, with consequences for scrapers and data collection (same batch).

Related Guides

• Assessing and Reducing Secondary Copyright Liability: A Practical Guide: the step-by-step companion for services and rightsholders.
• Digital Millennium Copyright Act Safe Harbors for Online Service Providers: safe harbor conditions in detail.
• Running a Platform That Hosts User Content: A Practical Guide: running operations for a hosting business.
• Identifying an Anonymous Defendant Online: A Practical Guide: finding the direct infringer.

Related Checklists

• Secondary Copyright Liability Risk Checklist: A Practical Checklist: an exposure review organized by theory.
• DMCA Safe Harbor Compliance Checklist: A Practical Checklist: conditions a provider must meet to keep the safe harbor.

Related Toolkits

• Secondary Liability Toolkit: Repeat Infringer Policies, Notice Workflows, and Inducement Audits: policies, workflows, audits, and demand templates.
• Platform Liability Toolkit: Terms of Service, Notice Procedures, and Litigation Defenses: terms and notice procedures for platforms.