Mavrix v. LiveJournal: Copyright

The Ninth Circuit held volunteer moderators may be a platform's agents, so posts they screen may fall outside DMCA 512(c)'s user-direction safe harbor.

Few decisions have unsettled the operators of moderated online communities like Mavrix Photographs, LLC v. LiveJournal, Inc., 873 F.3d 1045 (9th Cir. 2017). A celebrity photo agency sued a social media platform over twenty paparazzi photographs that users had submitted to the platform’s most popular fan community, and the Ninth Circuit held that the volunteer moderators who screened and approved those submissions might be the platform’s common-law agents. If they were, the photographs may not have been stored “at the direction of a user” at all, and the platform would lose the threshold protection of the Digital Millennium Copyright Act’s hosting safe harbor, 17 U.S.C. § 512(c).

The decision, written by Judge Richard A. Paez, reversed a summary judgment that LiveJournal had won below. In doing so it converted content moderation, an activity Congress plainly wanted to encourage, into a litigable question of agency law that every curated user-generated-content platform now has to analyze. Nearly a decade later, Mavrix remains the leading appellate authority on when human curation forfeits the safe harbor’s first requirement.

At a glance

Case: Mavrix Photographs, LLC v. LiveJournal, Inc., 873 F.3d 1045 (9th Cir. 2017), No. 14-56596, appeal from the Central District of California (Carney, J.)
Decided: Opinion filed April 7, 2017, and amended August 30, 2017; panel of Judges Pregerson, Paez, and Christen, opinion by Judge Paez; reversed, vacated, and remanded
Holding: Common-law agency applies to the section 512(c) safe harbor, so if a platform’s volunteer moderators are its agents, material they screen and publicly post may not be stored “at the direction of a user,” and genuine factual disputes about the moderators’ agency status precluded summary judgment for LiveJournal.
Status: Final as an appellate precedent; the amended opinion remains the controlling published word in the litigation, which produced no further published merits decision after remand.

The section 512(c) safe harbor and its user-direction threshold

Congress enacted the DMCA safe harbors in 1998 to balance the interests of copyright holders, technology entrepreneurs, and the public. Section 512 creates four safe harbors, and the one that matters for hosting platforms is section 512(c), which limits liability “for infringement of copyright by reason of the storage at the direction of a user” of material residing on the provider’s system.

The phrase “at the direction of a user” is the gate. Only if the provider clears it does the familiar machinery of the safe harbor come into play: the provider must lack actual and “red flag” knowledge of the specific infringement, must not receive a financial benefit directly attributable to infringing activity it has the right and ability to control, and must expeditiously remove material upon proper notice. 17 U.S.C. § 512(c)(1). The Ninth Circuit had elaborated those elements in UMG Recordings, Inc. v. Shelter Capital Partners LLC, 718 F.3d 1006 (9th Cir. 2013), and the Second Circuit had done the same in Viacom International, Inc. v. YouTube, Inc., 676 F.3d 19 (2d Cir. 2012). Both courts had also blessed “accessibility-enhancing” activities such as automated reformatting of user uploads, and the Fourth Circuit in CoStar Group, Inc. v. LoopNet, Inc., 373 F.3d 544 (4th Cir. 2004), had extended that tolerance to a cursory manual screen of photographs.

What no appellate court had squarely decided was the question Mavrix presented: what happens when human beings, organized and directed by the platform, substantively review every submission and publicly post only the ones they select?

Oh No They Didn’t: the facts and the posture

LiveJournal is a social media platform built around thematic “communities” that users create and run under their own rules. Its runaway success story was Oh No They Didn’t (ONTD), a celebrity news community that by 2010 was drawing 52 million page views a month. Users submitted proposed posts containing photographs, videos, links, and gossip to an internal queue. A team of nine moderators, six maintainers, and one owner reviewed the submissions and publicly posted roughly one-third of them, screening for relevance to “new and exciting celebrity news” as well as for copyright infringement, pornography, and harassment.

In 2010, LiveJournal decided to capture advertising revenue from ONTD’s popularity. It hired an active moderator, Brendan Delzer, as the community’s full-time “primary leader,” intending to “take over” ONTD, grow it, and run ads on it. Delzer instructed the volunteer moderators on what content to approve and selected and removed moderators based on performance, while continuing to review posts himself.

Mavrix Photographs is a celebrity photo agency specializing in candid shots of celebrities in tropical locations; its business depends on selling exclusive images, such as photographs breaking the news of Beyoncé’s pregnancy, to celebrity magazines. Between 2010 and 2014, ONTD publicly posted twenty Mavrix photographs in seven posts, some bearing a generic watermark and some a watermark reading “Mavrixonline.com.” Mavrix never sent a DMCA takedown notice. It sued instead, and LiveJournal removed the posts when the suit was filed. The district court granted LiveJournal summary judgment, holding that the safe harbor applied because users had submitted the photographs, and it separately refused to let Mavrix discover the anonymous moderators’ identities.

Agency law comes to the DMCA

The Ninth Circuit reversed. The district court, Judge Paez wrote, had focused on the wrong act: users’ submission of photographs describes the passive receipt of material that a different safe harbor, section 512(a), protects, while section 512(c) focuses on the provider’s role in publicly posting infringing material on its site. Posting, not submission, is the critical inquiry, and LiveJournal posted the photographs only after its moderator team reviewed and approved them. Whether they were “truly stored at the direction of the user” therefore depended on whether the moderators’ acts could be attributed to LiveJournal.

That attribution question, the court held, is governed by the common law of agency. Statutes are presumed not to displace the common law absent clear language, and both the Supreme Court, in Community for Creative Non-Violence v. Reid, 490 U.S. 730 (1989), and the Ninth Circuit itself had applied common-law principles in copyright and DMCA cases. Applying the Restatement (Third) of Agency, the panel found genuine disputes of material fact on every front. On actual authority, Mavrix presented evidence that LiveJournal gave moderators explicit screening directions and relied on their work as “an integral part of its screening and posting business model”; that the moderators were unpaid did not matter, because agents “need not receive payment from their principal.” On apparent authority, at least one user whose post was removed complained to LiveJournal that “the moderators checked and approved my post,” treating moderator approval as the platform’s own act. And on control, Delzer supervised the moderators substantively and removed them for poor performance, while the moderators remained “free to leave and go and volunteer their time in any way they see fit.” A reasonable jury, the court concluded, could find an agency relationship, so summary judgment was improper.

Critically, the panel added that even an agency finding would not end the case. The fact finder would still have to ask whether the moderators’ work was merely “accessibility-enhancing,” the kind of screening Shelter Capital and CoStar tolerated, or whether their “extensive, manual, and substantive activities” in approving roughly a third of submissions for celebrity-gossip value went beyond it. The court also vacated the discovery order shielding the moderators’ identities, since the agency analysis may turn on who they were and what they did.

Guidance on knowledge, financial benefit, and the August amendment

Because the issues would recur on remand, the panel went on to construe the remaining safe harbor elements. On knowledge, it stressed that Mavrix’s failure to send a takedown notice “strip[s] it of the most powerful evidence of [actual] knowledge,” but that this evidence is powerful rather than conclusive: the fact finder must also assess the provider’s subjective knowledge, including what agents like the moderators knew. On red flag knowledge, the question is whether the provider was aware of facts making the specific infringement “objectively obvious to a reasonable person,” and here the watermarks matter: the fact finder should ask whether it would be objectively obvious that photographs bearing a generic watermark or a “Mavrixonline.com” watermark were infringing. On financial benefit, the court reiterated the Shelter Capital rule that the “right and ability to control” requires “something more” than the power to remove or block posts, and noted Mavrix’s contested evidence that roughly 84 percent of ONTD posts contained infringing material on a community LiveJournal monetized through advertising.

The original April opinion drew heavy criticism for counting LiveJournal’s list of complained-about sources and its automatic blocking tool as evidence of control, a framing commentators warned would punish anti-piracy tooling. On August 30, 2017, the panel amended the opinion, published at 873 F.3d 1045, removing that language and clarifying that a platform’s use of automatic blocking software does not weigh against safe harbor eligibility.

Open questions

Mavrix left the hardest lines undrawn. The court never said how much substantive curation converts screening from accessibility-enhancing activity into platform-directed posting; it identified a spectrum running from automated reformatting through CoStar’s cursory gatekeeping to ONTD’s editorial selection, and left the fact finder to place any given moderation program on it. It likewise did not decide when volunteers actually are agents, only that a jury could so find on this record, so the level of instruction, supervision, and removal authority a platform can safely exercise remains unquantified. The interaction with section 512(m), which promises that monitoring for infringement will not itself create liability, is still uneasy: Mavrix preserves the promise formally while making organized human review the very fact that can forfeit the threshold requirement. Because the litigation produced no further published decision after remand, none of these questions received a definitive answer in the case itself; as of July 2026 the amended opinion stands as the controlling guidance in the Ninth Circuit.

Implications for creators and businesses

Structured moderation is a legal design choice, not just a product one. A platform that hands volunteers screening criteria, supervises them through employees, and removes them for poor performance is building the factual record for an agency finding. Platforms that want the safe harbor should document moderator independence or confine pre-publication review to narrow, mechanical checks.
Watermarks do real work for photographers. The panel treated a visible watermark, especially one naming the rights holder’s website, as evidence bearing on red flag knowledge. Watermarking strengthens both the infringement case and the answer to a safe harbor defense.
Skipping takedown notices has costs. Mavrix’s decision not to use LiveJournal’s DMCA process deprived it of the best evidence of the platform’s actual knowledge. Rights holders planning litigation should still paper the record with compliant section 512(c)(3) notices.
Anti-piracy tooling is safe; editorial curation is the risk. After the August 2017 amendment, automatic blocking software and infringement lists do not count against a platform. What jeopardizes the safe harbor is substantive human selection of what gets published, particularly when it serves the platform’s advertising business.

Frequently asked questions

Did the Ninth Circuit hold that LiveJournal lost the DMCA safe harbor? No. The court held only that LiveJournal could not win the safe harbor on summary judgment. Genuine disputes of material fact about whether the ONTD moderators were LiveJournal’s common-law agents required a remand for trial, where a fact finder would resolve agency and, depending on the answer, the remaining elements of the section 512(c) defense.

Does using volunteer moderators automatically make them a platform’s agents? No. Agency is a fact-specific inquiry into authority and control. The Mavrix panel pointed to evidence that LiveJournal gave moderators express screening criteria, supervised them through a paid employee who removed them based on performance, and relied on their screening as part of its business model. Volunteers who operate independently, without that direction and supervision, stand on very different footing.

What is the difference between actual knowledge and red flag knowledge under section 512(c)? Actual knowledge is subjective: did the provider actually know of the specific infringement. Red flag knowledge is objective: was the provider aware of facts that would have made the specific infringement obvious to a reasonable person. In Mavrix, the panel instructed the fact finder to ask whether it would be objectively obvious to a reasonable person that photographs bearing a generic watermark or a Mavrixonline.com watermark were infringing.