On Wednesday, the EU will vote on whether to accept two controversial proposals in the new Copyright Directive; one of these clauses, Article 13, has the potential to allow anyone, anywhere in the world, to effect mass, rolling waves of censorship across the Internet.
The way things stand today, companies that let their users communicate in public (by posting videos, text, images, etc) are required to respond to claims of copyright infringement by removing their users' posts, unless the user steps up to contest the notice. Sites can choose not to remove work if they think the copyright claims are bogus, but if they do, they can be sued for copyright infringement (in the United States at least), alongside their users, with huge penalties at stake. Given that risk, the companies usually do not take a stand to defend user speech, and many users are too afraid to stand up for their own speech because they face bankruptcy if a court disagrees with their assessment of the law.
This system, embodied in the United States' Digital Millennium Copyright Act (DMCA) and exported to many countries around the world, is called "notice and takedown," and it offers rightsholders the ability to unilaterally censor the Internet on their say-so, without any evidence or judicial oversight. This is an extraordinary privilege without precedent in the world of physical copyright infringement (you can't walk into a cinema, point at the screen, declare "I own that," and get the movie shut down!).
But rightsholders have never been happy with notice and takedown. Because works that are taken down can be reposted, sometimes by bots that automate the process, rightsholders have called notice and takedown a game of whac-a-mole, where they have to keep circling back to remove the same infringing files over and over.
Rightsholders have long demanded a "notice and staydown" regime. In this system, rightsholders send online platforms digital copies of their whole catalogs; the platforms then build "copyright filters" that compare everything a user wants to post to this database of known copyrights, and block anything that seems to be a match.
Tech companies have voluntarily built versions of this system. The most well-known of the bunch is YouTube's Content ID system, which cost $60,000,000 to build, and which works by filtering the audio tracks of videos to categorise them. Rightsholders are adamant that Content ID doesn't work nearly well enough, missing all kinds of copyrighted works, while YouTube users report rampant overmatching, in which legitimate works are censored by spurious copyright claims: NASA gets blocked from posting its own Mars rover footage; classical pianists are blocked from posting their own performances, birdsong results in videos being censored, entire academic conferences lose their presenters' audio because the hall they rented played music at the lunch-break—you can't even post silence without triggering copyright enforcement. Besides that, there is no bot that can judge whether something that does use copyrighted material is fair dealing. Fair dealing is protected under the law, but not under Content ID.
If Content ID is a prototype, it needs to go back to the drawing board. It overblocks (catching all kinds of legitimate media) and underblocks (missing stuff that infuriates the big entertainment companies). It is expensive, balky, and ineffective.
It's coming soon to an Internet near you.
On Wednesday, the EU will vote on whether the next Copyright Directive will include "Article 13," which makes Content-ID-style filters mandatory for the whole Internet, and not just for the soundtracks of videos—also for the video portions, for audio, for still images, for code, even for text. Under Article 13, the services we use to communicate with one another will have to accept copyright claims from all comers, and block anything that they believe to be a match.
This measure will will censor the Internet and it won't even help artists to get paid.
Let's consider how a filter like this would have to work. First of all, it would have to accept bulk submissions. Disney and Universal (not to mention scientific publishers, stock art companies, real-estate brokers, etc) will not pay an army of data-entry clerks to manually enter their vast catalogues of copyrighted works, one at a time, into dozens or hundreds of platforms' filters. For these filters to have a hope of achieving their stated purpose, they will have to accept thousands of entries at once—far more than any human moderator could review.
But even if the platforms could hire, say, 20 percent of the European workforce to do nothing but review copyright database entries, this would not be acceptable to rightsholders. Not because those workers could not be trained to accurately determine what was, and was not, a legitimate claim—but because the time it would take for them to review these claims would be absolutely unacceptable to rightsholders.
It's an article of faith among rightsholders that the majority of sales take place immediately after a work is released, and that therefore infringing copies are most damaging when they're available at the same time as a new work is released (they're even more worried about pre-release leaks).
If Disney has a new blockbuster that's leaked onto the Internet the day it hits cinemas, they want to pull those copies down in seconds, not after precious days have trickled past while a human moderator plods through a queue of copyright claims from all over the Internet.
Combine these three facts:
1. Anyone can add anything to the blacklist of "copyrighted works" that can't be published by Internet users;
2. The blacklists have to accept thousands of works at once; and
3. New entries to the blacklist have to go into effect instantaneously.
It doesn't take a technical expert to see how ripe for abuse this system is. Bad actors could use armies to bots to block millions of works at a go (for example, jerks could use bots to bombard the databases with claims of ownership over the collected works of Shakespeare, adding them to the blacklists faster than they could possibly be removed by human moderators, making it impossible to quote Shakespeare online).
These entities couldn't use Content ID to censor the whole Internet: instead, they had to manually file takedowns and chase their critics around the Internet. Content ID only works for YouTube — plus it only allows "trusted rightsholders" to add works wholesale to the notice and staydown database, so petty censors are stuck committing retail copyfraud.
But under Article 13, everyone gets to play wholesale censor, and every service has to obey their demands: just sign up for a "rightsholder" account on a platform and start telling it what may and may not be posted. Article 13 has no teeth for stopping this from happening: and in any event, if you get kicked off the service, you can just pop up under a new identity and start again.
Some rightsholder lobbyists have admitted that there is potential for abuse here, they insist that it will all be worth it, because it will "get artists paid." Unfortunately, this is also not true.
For all that these filters are prone to overblocking and ripe for abuse, they are actually not very effective against someone who actually wants to defeat them.
Let's look at the most difficult-to-crack content filters in the world: the censoring filters used by the Chinese government to suppress "politically sensitive" materials. These filters have a much easier job than the ones European companies will have to implement: they only filter a comparatively small number of items, and they are built with effectively unlimited budgets, subsidized by the government of one of the world's largest economies, which is also home to tens of millions of skilled technical people, and anyone seeking to subvert these censorship systems is subject to relentless surveillance and risks long imprisonment and even torture for their trouble.
Those Chinese censorship systems are really, really easy to break, as researchers from the University of Toronto's Citizen Lab demonstrated in a detailed research report released a few weeks ago.
People who want to break the filters and infringe copyright will face little difficulty. The many people who want to stay on the right side of the copyright —but find themselves inadvertently on the wrong side of the filters—will find themselves in insurmountable trouble, begging for appeal from a tech giant whose help systems all dead-end in brick walls. And any attempt to tighten the filters to catch these infringers, will of course, make it more likely that they will block non-infringing content.
A system that allows both censors and infringers to run rampant while stopping legitimate discourse is bad enough, but it gets worse for artists.
Content ID cost $60,000,000 and does a tiny fraction of what the Article 13 filters must do. When operating an online platform in the EU requires a few hundred million in copyright filtering technology, the competitive landscape gets a lot more bare. Certainly, none of the smaller EU competitors to the US tech giants can afford this.
On the other hand, US tech giants can afford this (indeed, have pioneered copyright filters as a solution, even as groups like EFF protested it), and while their first preference is definitely to escape regulation altogether, paying a few hundred million to freeze out all possible competition is a pretty good deal for them.
The big entertainment companies may be happy with a deal that sells a perpetual Internet Domination License to US tech giants for a bit of money thrown their way, but that will not translate into gains for artists. The fewer competitors there are for the publication, promotion, distribution and sale of creative works, the smaller the share will be that goes to creators.
We can do better: if the problem is monopolistic platforms (and indeed, monopolistic distributors), tackling that directly as a matter of EU competition law would stop those companies from abusing their market power to squeeze creators. Copyright filters are the opposite of antitrust, though: it will make the biggest companies much bigger, to the great detriment of all the "little guys" in the entertainment industry and in the market for online platforms for speech.