Cox Loses DMCA Safe Harbor but Gets a New Trial on Contributory Infringement–BMG v. Cox
The first notice alleging a subscriber’s infringement produces no action from Cox. The second through seventh notices result in warning emails from Cox to the subscriber. After the eighth and ninth notices, Cox limits the subscriber’s Internet access to a single webpage that contains a warning, but the subscriber can reactivate complete service by clicking an acknowledgement. After the tenth and eleventh notices, Cox suspends services, requiring the subscriber to call a technician, who, after explaining the reason for suspension and advising removal of infringing content, reactivates service. After the twelfth notice, the subscriber is suspended and directed to a specialized technician, who, after another warning to cease infringing conduct, reactivates service. After the thirteenth notice, the subscriber is again suspended, and, for the first time, considered for termination. Cox never automatically terminates a subscriber.
Cox also limited its processing of notes from copyright owners or agents.
BMG hired Rightscorp to monitor police infringements. Rightscorp issues takedown requests, and it asks ISPs to forward settlement requests to the subscribers. Prior to BMG having engaged Rightscorp, Comcast decided to cease processing takedown requests from Rightscorp. Rightscorp sent “millions of notices” to Cox on BMG’s behalf, but BMG never viewed any of these.
The trial court held that a reasonable jury could not conclude that Cox maintained a policy of terminating repeat infringers. BMG put forth evidence from which a jury could conclude that Cox knew of accounts being used to effect infringing activity but nevertheless failed to terminate such accounts. The case went to trial (essentially) on damages, and the verdict came back for $25MM in BMG’s favor.
Cox is Not Entitled to the DMCA Safe Harbor: Cox argued that the repeat infringer policy of 512(i)(1)(A) obligates a service provider to terminate those “who have been held liable by a court for multiple instances of copyright infringement.” The court rejects this argument based on the structure of the statute and its legislative history. Where necessary, the drafters showed they could distinguish between alleged or apparent infringement and activity which is “ultimately determined [by a court]” to be infringing. The statute did not employ the latter definition.
The court then says Cox did not implement a repeat infringer policy:
- Numerous emails showed that Cox’s termination of subscribers was symbolic (and followed by indiscriminate reactivation).
- By disregarding all notices from Rightscorp, Cox made no effort to terminate users to infringed on BMG’s rights in particular.
- There was evidence that Cox actually reactivated (or failed to terminate) subscribes Cox subjectively thought were infringing.
Cox argued that the evidence in question did not show actual knowledge of repeat infringement (of BMG’s rights), but the court says this is misplaced. Cox failed to meet its burden of showing it had a reasonable policy.
Standards for Contributory Infringement: The court’s instruction allowed the jury to impose liability if the jury found “Cox knew or should have known of [the] infringing activity.” The court says this was an incorrect statement of the law. As an initial matter, Cox argued that it should be absolved from liability where its technology is capable of “substantial noninfringing uses.” The court rejects this argument.
The court evaluates a range of cases, including Grokster, and several 9th Circuit cases. It says the cases appear to agree that intent and knowledge (or its equivalent, willful blindness) suffice to establish liability, but the cases are less clear on lesser standards. Ultimately, the court says that lesser mental states do not support a finding of contributory liability. BMG cited to the 9th Circuit’s Ellison decision, but the court says the Ninth Circuit “clarified” this ruling in Luvdarts v. ATT.
The court agrees with Cox on two other disputes regarding jury instruction. First, the court says that the trial court’s instruction allowed the jury to impute knowledge based on Cox’s “generalized knowledge” of infringing activity. The court says that the instruction should require a finding of knowledge as to specific customers. Second, the court says the willful blindness instruction was also broad:
[s]ince we have held that contributory infringement requires knowledge of, or willful blindness to, specific instances of infringement, the court’s willful blindness instruction should similarly require a conclusion that Cox consciously avoid learning about specific instances of infringement, not merely that Cox avoided confirming the fact that ‘Cox users were infringing on BMG’s copyrights’ in general.
The court concludes that BMG offered “powerful evidence” of willful blindness, but a properly instructed jury should make that determination.
The court rejects some other Cox arguments, including that the Rightscorp takedowns were hearsay. Hearsay only applies to statements by people—computer-generated statements don’t have a hearsay problem.
This seems like a win, of sorts, for platforms, although not a great result for Cox specifically. The court rejects a less-than-knowledge liability standard, and says knowledge or its equivalent must be with respect to specific instances of infringement. Rampant instances of infringement occurring somewhere on the platform does not trigger any obligation to police the platform or its users. This is consistent with appeals court rulings from the Second and Ninth Circuits. (It was interesting to see the court here cite to Luvdarts v. AT&T, a derivative liability case that did not receive much attention.)
The court doesn’t say much about the number of chances Cox gives its users. One wonders if the policy had been properly enforced, whether the court would have said 12 chances is too many to constitute an effective repeat infringer policy. The court also doesn’t spend much time on what looks like a key fact: Cox’s decision to stop honoring Rightscorp takedown notices. It receives a few brief mentions, but it would have been nice to hear what the court thought about it. Platforms often force content owners to jump through (perhaps minor) hoops to submit takedown notices (e.g., registration; use of an online interface). Here, Cox went to the extreme, and lost the safe harbor entirely.
Finally, this comment from Mike Masnick is worth reading: “Appeals Court Makes A Mess Of Copyright Law Concerning ISPs And Safe Harbors“. It raises an interesting point about whether Cox had an obligation to terminate users at all, since it was only an access provider rather than a host. Cox did not appear to squarely raise this argument on appeal, and having set up a “policy” it was stuck with having to enforce it (and was held liable for grossly falling short).
This is a frustrating case because Cox made several dubious decisions that were tough to defend in court:
- Cox’s 13 strike policy, with a max of one-strike-per-day and 6 month reset, is baffling. I don’t if any other industry player has such a lackadaisical attitude about recidivists.
- Cox thought the statutory termination requirement meant “of that account,” not “of that subscriber.” Thus, Cox would freely reactivate “terminated” subscribers, apparently for the revenues. For example, the court recites this evidence from the head of Cox’s abuse team: “Once the customer has been terminated for DMCA, we have fulfilled the obligation of the DMCA safe harbor and can start over [so Cox could] collect a few extra weeks of payments for their account. ;-).” [EMOTICON ALERT!] The court responds: “An ISP cannot claim the protections of the DMCA safe harbor provisions merely by terminating customers as a symbolic gesture before indiscriminately reactivating them within a short timeframe.” Then again, Cox’s focus on revenue highlights the ill-fitting nature of deputizing IAPs as copyright cops. IAPs prioritize their own revenue, not the concerns of copyright owners or protecting legitimate customers from overreaching copyright demands.
- When Cox putatively changed its policy to terminate subscribers, it implicitly stripped termination from the 13-strike policy, so it still virtually never terminated subscribers.
- Cox blacklisted Rightscorp as a submitter of DMCA notices. The court responds: “Cox’s decision to categorically disregard all notices from Rightscorp provides further evidence that Cox did not reasonably implement a repeat infringer policy.” The court doesn’t discuss why Cox made that move, but presumably it has something to do with low-quality takedown notices. But perhaps the solution should lie within 512(f), not Cox putting its fingers in its ears and saying “I’m not listening…” I’d love to hear from Cox’s lawyer who blessed that approach.
- Cox’s position that “repeat infringer” only means “adjudicated repeat infringer” is not credible. Although Cox has a point about takedown abuse, the statutory remedy is supposed to be in 512(f), not in raising the bar on copyright owners’ ability to send takedown notices. I find it hard to believe that any lawyer blessed that statutory interpretation. Honestly, this argument feels like the kind of thing that defense litigators make up after a lawsuit has commenced, and only because all other better arguments aren’t available.
The court’s discussion of contributory copyright infringement’s scienter requirements is quite interesting. First, the court reinforces that the strong form of Sony’s rule–no liability for any device “capable of substantial non-infringing uses”–probably does not protect online services. Grokster severely undermined the Sony rule, and here the court thumps Cox’s attempt to invoke it. However, the court does preserve the Sony rule in one respect:
Selling a product with both lawful and unlawful uses suggests an intent to cause infringement only if the seller knows of specific instances of infringement, but not if the seller only generally knows of infringement.
Second, the court says willful blindness can substitute for contributory infringement’s knowledge requirement, but the court explicitly indicates that willful blindness is a more rigorous scienter standard than recklessness. This makes at least three circuits that seemingly accept a willful blindness exception to the DMCA’s safe harbors, along with the Second and Ninth circuits.
Third, the opinion provides a definition of “willful blindness” that’s far more coherent than what the Second or Ninth Circuit’s non-definitions. The court says willful blindness occurred if “Cox consciously avoided learning about specific instances of infringement, not merely that Cox avoided confirming the fact that ‘Cox users were infringing BMG’s copyrights’ in general.” For now, given the relative clarity of this articulation, I’m going to say that this is the best prevailing definition of willful blindness in the DMCA/contributory copyright infringement context: “the defendant consciously avoided learning about specific instances of infringement.”
Fourth, the court says negligence/”should have known” does not satisfy the knowledge requirement of contributory infringement, which is the putative defense win that Venkat identifies. Thus, it’s not good enough for the copyright owner to show that “Cox knew of ‘direct infringement of BMG’s copyrighted works’ by its subscribers if Cox had data showing that some number of its subscribers were infringing BMG’s copyrights…if the data did not show which ones were infringing. So Cox will get a new trial where “the contributory infringement instruction should require that Cox knew of specific instances of infringement or was willfully blind to such instances.” I’m not sure how much of a win this is for Cox. After all, Cox’s functionally automatic reactivation of recidivist subscribers, and its categorical refusal to consider Rightscorp’s takedown notices, place Cox in a difficult position.
I’ll defer to evidence experts about the hearsay ruling that “the fact that the machine-generated notices also contained the signature of Rightscorp’s CEO and an oath under penalty of perjury does not transform them into statements, since the information itself was not prepared or created by a human.” This is a baffling statement to someone who isn’t an evidence expert. The automated notices were configured by humans, and any textual parts of the notice were written by humans. If this principle applied outside the context of evidence law, it would raise powerful questions about whether AI-created works are copyrightable or search engines have First Amendment-protectable speech rights. However, this is just a niche aspect of evidence law, so any techno-haters hoping to recycle this passage more broadly should calm down.
Case citation: BMG Rights Mgmt (US) LLC v. Cox Communications Inc., 2018 WL 650316 (4th Cir. Feb. 1, 2018).