At the end of 2021, YouTube’s first Copyright Transparency Report 2021 (“Report”) was published. It is interesting to look at this Report against the background of the 2019 EU rules for the liability of platforms like YouTube through the famous Art. 17 DSM Directive 2019/790 (“DSMD”). But first let’s take a look at Mars (the planet).
In 2012, US NASA’s robotic rover touched down on Mars. NASA uploaded a video of this historic event onto YouTube. It was taken down after one hour, as it was subject to a copyright notice by a news channel relying on the US DMCA. This news channel had used the (open source) NASA video for its own news video and ended up raising unjustified claims against other copies on YouTube, including NASA’s original video. A commentator noted that “NASA’s new rover arrives on Mars, crash lands in YouTube’s DMCA hell”. Another – more recent – example involves the video of a 2020 copyright panel at New York University where music was played to explain the application of the US fair use rules. The video was caught by a takedown – but ended up being put back online on YouTube.
These and other well-known examples tend to create the public impression that abusive or otherwise unjustified copyright takedown requests are the general rule on YouTube. But is this really true? To answer this question, it seems important not to rely on illustrative single stories. “We are naturally drawn to extreme examples, and they are so easy to recall”, says Hans Rosling in his book Factfulness (2018, p. 42). Rather, it is necessary “to learn about the world as it really is”, to quote Rosling again (p. 255).
Regarding the question of unjustified copyright takedown on YouTube, YouTube’s Report provides a chance to take a fact-based or “factful” look – assuming that YouTube has correctly indicated the facts. It is the first ever report by YouTube on copyright actions. It only covers half the year from January to June 2021, which is the period before Art. 17 DSMD was implemented into the national laws of the EU member states. The Report has already been discussed by Paul Keller on the Kluwer Copyright Blog – here are my main takeaways.
- Illegal uploads remain a massive problem on YouTube
The Report identifies hundreds of millions of copyright actions through the different YouTube copyright tools by rightholders: ContentID had ca. 722 million, while the Enterprise Webform (ca. 3 million), Copyright Match (ca. 1.5 million) and the Webform (ca. 4 million) account for far fewer copyright actions (page 5). The sheer number is striking: We are talking about ca. 2 million copyright actions per day. These numbers have to be put into context relative to the disputes raised over such requests. Only 0.5% of all copyright actions via ContentID (i.e. 3.7 million) were disputed. The figures for copyright actions via tools other than ContentID were higher, but still covered only a minor share: Copyright Match 1.3%, Enterprise Webform 1.9% and Webform 5.2%. It is likely that these numbers only reflect a certain share of unjustified copyright actions, because research suggests that uploaders may not be willing to stand up and complain to YouTube (Paul Keller op cit., citing Jennifer M. Urban, Joe Karaganis, Brianna Schofield, Notice and Takedown in Everyday Practice; the underlying data is, however, from 2017). Yet, it seems to be a fact that hundreds of millions of uploads on YouTube were copyright infringing. Consequently, platforms like YouTube need effective rules to motivate them to fight copyright infringements.
Art. 17 DSMD tries to establish such rules. Under that provision, as a general rule, platforms like YouTube are liable for the making available of infringing content on their platforms, unless they comply with certain duties (see Art. 17 (4) DSMD). This starting point seems to be justified looking at the facts of the Report.
- Unjustified takedowns? Need to differentiate between notifiers
Copyright actions on such large and important platforms like YouTube may be abusive or otherwise unjustified. The underlying reasons may vary, as mentioned in the Report: For example, “lack of understanding of either copyright law or our tools” (p. 6) and intentional abuse (p. 6). Examples of such abuse are attempts by political actors to censor political speeches or companies trying to stifle criticism of their products or services (p. 6). But the facts indicated in the Report show that the share of unjustified claims depends to a large extent on the rightholder (notifier) initiating the copyright action (page 8 et seq.):
- ContentID: Only 0.5% of the requests were disputed by uploaders. And only some 60% of this small share of counter-notices were actually successful. As the Report sets out, rightholders allowed to use ContentID have to meet qualifying criteria: A need for a scaled solution, working knowledge in copyright and the necessary resources to manage a complex tool (p. 10). In practice, only movie studios, record labels, collecting societies and a number of specialized service providers matched these criteria (p. 3). YouTube claims to enter into agreements with these “ContentID partners” (p. 3).
- Webforms: Numbers are quite different for users of the Webforms. For the simple Webform, open to all registered YouTube users, only 83.8% of the removal requests were processed by YouTube – YouTube rejected 8.6% as abusive and 6.52% as invalid. The percentage of counter-notifications received by YouTube (5.2%) was 10 times higher than following ContentID removals (0.5%). Interestingly, the numbers look quite different for the Enterprise Webform: 99.3% of the removal requests were accepted, no relevant abuses (0.0%) and only 0.5% of requests were invalid. The Enterprise Webform is open only to qualified rightholders, which likely explains the stark contrast in numbers.
What can we learn from this? To my mind, not that “overblocking is real” (Paul Keller op cit.). While this is theoretically correct, it does not tell the full story. The full story seems to be: The rate of unjustified copyright actions appears low – looking at ContentID it is very low – even though a degree of uncertainty remains with respect to potentially unjustified copyright actions with respect to which no counternotices were sent (see above 1.). Against this background, it seems justified to provide uploaders with an expeditious and effective complaint mechanism, subject to human review in case they think their upload was removed without justification. This is provided for example by Art. 17 (9) DSMD. That said, the Report does not provide any factual reasons to dramatize the possible extent of unjustified blocking.
Unjustified copyright requests have a strong correlation with the level of qualification and effort of the rightholder. This would support the model of the trusted flaggers which could gain more importance. This model would need to include non-discriminatory access to such a status for example on YouTube. The German implementation of Art. 17 DSMD already provides (to a limited extent) privileges to trusted flaggers (Section 14 (4) German Copyright Service Provider Act – UrhDaG). Also, the European Commission communication on Art. 17, 2021 (Communication Art. 17 DSMD) recognizes that Online content-sharing service providers caught by the provision could chose to provide different tools to different rightholders (p. 14). In any case, 0.5% dispute rate for ContentID removals somewhat stands in interesting contrast to the principal criticism automated content recognition (ACR) technologies have received during the legislative process of Article 17 DSMD for alleged overblocking.
- There seems to be no alternative to automated removal – the “how” remains the major question
The Report shows absolute numbers of copyright claims that make you feel dizzy: Ca. 725 million requests through ContentID, and another ca. 6.5 million through the other copyright tools provided by YouTube in the six months from January to June 2021. There is no way to check all such requests or even a substantial proportion of them manually. Automated filtering seems unavoidable (see also Jonathan Baily, 7 Takeaways From YouTube’s Copyright Report ). While there does not seem to be a serious question about “whether” automated filtering should be used, the major question remains “how” you set its limits. Regarding Art. 17 DSMD, the proportionality principle in Art. 17 (5) DSMD needs to be filled with substance.
- The European Commission thinks that filtering should be limited in particular to manifestly infringing uploads (Communication Art. 17 DSMD, pp. 20 et seq.).
- The German implementation excepts certain “presumed legal uses” from the application of automated filtering. Such “presumed legal uses” have to remain public until the end of a successful complaint procedure, which has to be initiated by the rightholder. A “presumed legal use” requires a content-mix (mash-up) with less than 50% of the content owned by the requesting rightholder and it further requires that the content is either flagged by the uploader as legal or it is a minor use (Section 9 German Copyright Service Provider Act). A minor use would be for example 15 seconds of a film or 15 seconds of music (Section 10 German Copyright Service Provider Act). (On this blog, a future post by Julian Waiblinger and Jonathan Pukas will explore this German implementation further.)
This is not the forum to discuss and assess such solutions. The Report does not contain any data to support or reject them.
- Art. 17 DSMD sets an international standard for YouTube
The Report indicated that by October 2021 YouTube also allowed Webform users to request automated filtering – and not only mere takedown (p. 2). While the instruction to block still seems to work on a manual basis for Webform users, the shift to open filtering up for all rightholders is likely caused by YouTube’s duties under Art. 17 (4) b and c DSMD. After the GDPR, this is another example showing that EU law may have an impact on the legal set up of companies operating internationally like YouTube. Also, Art. 17 DSMD would then have the so-called “Brussels effect” (Paul Keller, op cit.).
- Outlook – hopefully more facts
Unfortunately, the Report does not provide any regional breakdown of the facts. Also, it does not cover the time after the implementation of Art. 17 DSMD in the EU member states. In the future, YouTube may be obliged – like other online content-sharing service providers – to provide transparency reports on their work for the purposes of stakeholder dialogues (Article 17 (10) DSMD). The Commission also addresses this issue in its Communication on Art. 17 (p. 23).
One problem, however, remains: Besides the uploaders and the rightholders, YouTube itself is a key stakeholder in the triangle of interests in copyright requests. YouTube is not neutral. It cannot be ruled out that YouTube may try to select the facts published to push the public assessment of Art. 17 DSMD in a certain direction. This brings up the question of third party access to YouTube’s data. Interestingly, the German legislator has tried to provide a solution for this: Research organizations, which come under the exception for text and data mining for the purposes of scientific research (Art. 3 DSMD) also have a right to claim the data relevant for Art. 17 DSMD from YouTube and the like for the purpose of scientific research, insofar as this does not conflict with overriding justified interests of the platform. Additionally, the platform shall be entitled to reimbursement of costs incurred as a result in an appropriate amount.
In any case, this first YouTube Report is indeed a promising start (Paul Keller, op cit.), as it provides a first set of facts to “undramatize” copyright actions via YouTube and the like and start to look at this from a position of “factfulness”. To quote Hans Rosling once again: “I have found it useful and meaningful to learn about the world as it really is.” (Factfulness, p. 254). Following the Report this means: Copyright infringement needs to be addressed; unjustified requests have to be taken seriously, but seem to be an exceptional scenario in particular by qualified rightholders; and there is no real alternative to automation on larger platforms.