Content owners: Don’t Go Crazy with takedown notices.
When Prince’s record label send a Digital Millennium Copyright Act (DMCA) takedown notice to YouTube over a clip containing audio from “Let’s Go Crazy”, no one was surprised. What was surprising was that the person who uploaded the video pushed back. Eight years later, a victory in that fight will have a major impact on takedown notices and copyright enforcement online. The decision could provide some balance to an automated takedown system, restoring user rights and fair use.
Prince vs the internet
Prince has a troubled relationship with the internet. In 2007 he threatened to sue YouTube, eBay, and the Pirate Bay and hired a company called Web Sheriff to help get his music taken down. He sued 22 people for posting concert videos to YouTube in 2014 (he later dropped those claims). He famously announced “the internet’s completely over” in 2010. He purged his social media accounts in late 2014, soon after a Facebook Q&A in which he answered just one question about audio quality. In the summer of 2015, he removed his music from streaming services except Tidal.
Before the Web Sheriff rode into town, Prince’s online activities were handled by his record label, Universal Music. Universal had an assistant in their legal department search YouTube every day, looking for potentially infringing content.
Earlier that year, Stephanie Lenz had uploaded a video of her son to YouTube. The video featured the boy bouncing up and down to the sounds of Prince’s “Let’s Go Crazy”. It was cute and people liked it:
Universal’s legal assistant felt differently. After watching the video, he determined the song, not the baby, was “very much the focus of the video”. He signed off on a DMCA takedown notice to YouTube. The video came down on June 5, 2007.
Lenz filed a counter-notice claiming that the video was not infringing. Universal challenged the counter-notice on a technicality, pointing out that Lenz had not added in a declaration that her counter-notice was made under penalty of perjury. The video stayed down.
DMCA takedowns
Enforcement today has come a long way from the days of a lowly assistant searching Prince videos in a dark corner of the Universal offices. Now, content detection algorithms scour the internet for infringing content, sending automated DMCA takedowns when infringement is detected. The automated systems err on the side of over-enforcement, leading to countless stories of absurdly stupid DMCA abuse.
Even for the most ridiculous takedowns, internet companies will typically comply. The DMCA offers a powerful incentive to act quickly when copyright owners deliver a takedown notice. Refusal risks liability for users’ bad actions. The DMCA offers a “safe harbor” for providers who act quickly to take down infringing content—immunity from liability for users’ conduct.
The balance is supposed to come on the other side of the equation: the rightsholder sending the takedown notice only sending a notice if they have a “good faith belief” that the content is infringing. There is a general sense in the internet community that the balance has not worked out as planned. The Lenz case offered a chance to restore some balance.
Lenz Go Crazy
Normally the DMCA takedown story ends when the content is removed. Counter-notices are rare, and litigation is even rarer. The cost of litigation is simply too high, and the potential damages too low. YouTubers with little to gain and lots to lose would rather move on with their lives.
Lenz was an exception. She managed to secure pro bono representation from the Electronic Frontiers Foundation (EFF), and the fight was on. Lenz filed a claim against Universal. Her complaint wound its way through the courts, finally coming before the United States Court of Appeals for the 9th Circuit.
The court issued a lengthy decision in Lenz v Universal that will have significant implications for how takedowns are handled in the future.
The case depended on an interpretation of section 512(c)(3)(A)(v) of the DMCA, which says that a takedown notice must include “a statement that the complaining party has a goof faith belief that the use of the material in the manner complained of is not authorized by the copyright owner, its agents, or the law.”
The court had to decide whether fair use could be an authorization by the law. Lenz said yes. Universal said no. After a highly technical discussion of interpretation (see pages 11–15), the court sided with Lenz:
Fair use is not just excused by the law, it is wholly authorized by the law.
Next, the court had to decide if the takedown notice was sent with a “good faith belief” that the use was not authorized. Lenz said the takedown notice couldn’t have been sent in good faith since Universal didn’t even consider fair use, which could have authorized it by law.
The court held that fair use has to be considered before sending a takedown notice:
[F]air use is “authorized by the law” and a copyright holder must consider the existence of fair use before sending a takedown notification.
To be clear that this doesn’t mean copyright holders can just check a “No Fair Use” box on a takedown notice and carry on business as usual, the court continued:
A copyright holder who pays lip service to the consideration of fair use by claiming it formed a good faith belief when there is evidence to the contrary is still subject to [DMCA] liability.
Finally, the court found that even though Lenz hadn’t shown that she suffered monetary damages, she doesn’t have to. Instead, she could seek “nominal damages” in recognition of her rights, her expenses in bringing the case, and possibly even the EFF’s pro bono costs and legal fees.
What about takedowns?
While the Lenz saga will now go back to the lower court for a trial, this decision is already a big deal for user-generated content online. The takedown balance had been tipped in favor of content companies issuing automated takedowns, but the balance has been recalibrated.
From the EFF post on the decision:
Today, the United States Court of Appeals for the Ninth Circuit ruled that copyright holders like Universal must consider fair use before trying to remove content from the Internet. It also rejected Universal’s claim that a victim of takedown abuse cannot vindicate her rights if she cannot show actual monetary loss.
“Today’s ruling sends a strong message that copyright law does not authorize thoughtless censorship of lawful speech,” said EFF Legal Director Corynne McSherry. “We’re pleased that the court recognized that ignoring fair use rights makes content holders liable for damages.”
That leaves the question of how content owners can protect against actual infringement. When Universal’s legal assistants were manually searching YouTube in 2007, they were facing 10 hours of content being uploaded every minute. Today, it’s 400 hours per minute. There aren’t enough legal assistants in the world to watch it all.
The court acknowledged this issue, but made it clear that no matter how much content has to be sorted through, a fair use assessment is mandatory:
We are mindful of the pressing crush of voluminous infringing content that copyright holders face in a digital age. But that does not excuse a failure to comply with the procedures outlined by Congress.
[…]
Copyright holders cannot shirk their duty to consider — in good faith and prior to sending a takedown notification — whether allegedly infringing material constitutes fair use, a use which the DMCA plainly contemplates as authorized by the law. That this step imposes responsibility on copyright holders is not a reason for us to reject it.
What is a content company to do? Is this the case of a court out of touch with the realities of technology? Not this time. This time, the court was out in front of the technology, providing a roadmap for automated takedowns that consider fair use:
We note, without passing judgment, that the implementation of computer algorithms appears to be a valid and good faith middle group for processing a plethora of content while still meeting the DMCA’s requirements to somehow consider fair use.
[…]
[C]onsideration of fair use may be sufficient if copyright holders utilize computer programs that automatically identify for takedown notifications content where: “(1) the video track matches the video track of a copyrighted work submitted by a content owner; (2) the audio track matches the audio track of that same copyrighted work; and (3) nearly the entirety … is comprised of a single copyrighted work.
The court took no issue with automating takedown notices for direct copies of music videos, or when almost an entire video is “comprised of a single copyrighted work”. In fact, the decision suggests that some automated review for fair use in other cases would be acceptable as well, so long as a human is kept in the loop to make sure fair use is considered:
Copyright holders could then manually employ individuals like [Universal’s legal assistant] to review the minimal remaining content a computer program does not cull.
There are two immediate messages for content companies looking to send DMCA takedowns:
- If you are going to send a takedown, you have to take steps to consider fair use first. To send a takedown notice you have to affirm that the use of the work is not authorized by law. If there is fair use there is authorization.
- Fully automated takedowns are limited to perfect copies, or near perfect copies. If you want to send a takedown notice for a derivative work or a partial use of a work, you need to have a human review it first.
Racing with the machine
There will be grumbling and doomsday predictions from content companies whose automated takedown regimes are slowed by this decision, and from companies like Web Sheriff who will have to employ actual humans to generate takedowns.
These complaints should be ignored. While it adds time and expense to the takedown process, human review is a critical step on the path to a copyright balance. It is a great example of “racing with the machine” — humans and computers working together to achieve results neither could alone.
The problem also creates an opportunity for developers who can create algorithms to more accurately filter content for human review, all while still accounting for fair use. These algorithms will only get more valuable as the volume of user generated content continues to increase.
