An empirical study of DMCA takedown notices

Placeholder Show Content


This thesis is about copyright takedown notices -- the procedural mechanism introduced in the Digital Millennium Copyright Act which empowers copyright owners to submit notices with takedown requests to service providers, to have service providers expeditiously remove or disable access to allegedly infringing content. This is a novel mechanism that bypasses judicial oversight over copyright disputes in favor of a fast, cost-effective and efficient co-operative scheme for managing online infringement disputes. Since its inception in 1998, takedown notices have become the lynchpin of the DMCA. However, little is actually known about how they are actually used and by whom in practice. There is also little information about how reliable these notices are and whether mistakes are made. Even more importantly, there is no information about whether service providers have responded to each takedown request accurately and fairly. Conducting a census on half a million notices and more than 50 million requests between 2001 and 2012, the first paper of this thesis ("The State of the Discordant Union") finds that since 2008, there has been an exponential increase in the number of takedown notices submitted (from 4,387 notices/15,274 requests in 2009 to 443,911 notices/54,263,600 requests in 2012). Part of the reason for this increase was the Trusted Copyright Removal Program (TCRP) instituted by Google in April 2011, which permitted "reliable, high accuracy" reporters to electronically submit multiple takedown requests to Google for automated processing. This also made negligible the numbers of notices in non-web formats. The study also found that section 512(d) notices and requests directed to Google's Search services dominate (in 2012, at 94.4% of all notices and 99.9% of all requests). Finally, and rather surprisingly, the study found negligible numbers of counter-notices (89 counter-notices/0.02% of all notices in 2012). What about the quality of these notices? The second paper of this thesis ("Who watches the Watchmen?") finds that almost all notices comply with the non-functional formalities in the DMCA (signatures and attestations). However, 8.3% of all notices in 2012 fail to comply with the functional formalities (identification of the copyright work and the infringing material). In addition, at least 1.3% of takedown requests exhibit "substantive" errors that misidentify the copyright owner or provide inactive URIs as takedown requests. One of the most egregious examples is the continued submission of takedown requests directed to and similar non-functional cyberlocker sites. A total of 414,139 invalid takedown requests were issued, with every issuing reporter attesting to their accuracy. This was totally avoidable - eight of the top 30 reporters sent not a single one of these erroneous requests. To ensure that the takedown system remains efficient and error-free, the study proposes to strengthen the attestation requirements of notices, to require reporters to validate all takedown requests, and to subject recalcitrant reporters to the "slow lane" of a two-tier system for processing takedown notices. Google has however reported a rejection rate of (only) 2.5% of all takedown requests for 2012. Is Google's automated takedown system DMCA compliant? Using data analytics and machine learning techniques, the third paper ("Trust but Verify") attempts to model Google's takedown process and its relevant considerations for 55 million requests in 2012. These 20 considerations range from the statements of attestation to the identities of the reporters, and from the identities of the targeted domains to detailed information about the copyrighted work allegedly infringed. The resulting model (precision: 79.8%, recall: 84.0%, F1: 81.8%) is a reasonably accurate model of Google's takedown process. The model shows that Google's takedown process is largely DMCA compliant, particularly as regards its use of title information to ensure formal and substantive compliance. However, it exhibits some anomalies e.g. it complies with requests that have no descriptions of the copyrighted work, it largely ignores URI errors, and it accommodates TCRP members who have high rejection rates. The study observes that the strategic behavior of takedown systems to favor compliance to rejection is an undesirable by-product of the architecture of the DMCA. To correct these problems, the study suggests mandating the publication of reasons for each takedown request and empowering service providers with effective remedies against reporters who make knowing misrepresentations in their takedown requests. It is only through this that the overall quality of the service providers' takedown systems can be raised, and the right balance can be struck between protecting the interests of content providers in removing unauthorized online content and preserving the freedom of expression on the Internet.


Type of resource text
Form electronic; electronic resource; remote
Extent 1 online resource.
Publication date 2015
Issuance monographic
Language English


Associated with Seng, Daniel
Associated with Stanford University, School of Law JSD.
Primary advisor Lemley, Mark A, 1966-
Thesis advisor Lemley, Mark A, 1966-
Thesis advisor Hensler, Deborah R, 1942-
Thesis advisor Malone, Phillip R
Advisor Hensler, Deborah R, 1942-
Advisor Malone, Phillip R


Genre Theses

Bibliographic information

Statement of responsibility Daniel Seng.
Note Submitted to the School of Law JSD.
Thesis Thesis (JSD)--Stanford University, 2015.
Location electronic resource

Access conditions

© 2015 by Daniel Kiat Boon Seng
This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).

Also listed in

Loading usage metrics...