Last week, following riots that saw supporters of President Trump breach and sack parts of the Capitol building, Facebook and Twitter made the decision to give the president the boot. That was notable enough, given that both companies had previously treated the president, like other political leaders, as largely exempt from content moderation rules. Many of the president’s followers responded by moving to an alternative platform, Parler. This week, the response has taken a new turn by targeting. Infrastructure companies much closer to the bottom of the technical “stack” including Amazon Web Services (AWS), and Google’s Android and Apple’s iOS app storesdecided to cut off service to that alternative platform -- i.e., not just to an individual but to an entire site. Parler has so far struggled to return online, partly through errors of its own making, but also because the lower down the technical stack, the harder it is to find alternatives, or re-implement what capabilities the Internet has taken for granted.

Whatever you think of Parler, these decisions should give you pause. Private companies have strong legal rights under U.S. law to refuse to host or support speech they don’t like. But that refusal carries different risks when a group of companies comes together to ensure that forums for speech or speakers are effectively taken offline altogether.

The Free Speech Stack—aka “Free Speech Chokepoints”

To see the implications of censorship choices by deeper stack companies, let’s back up for a minute. As researcher Joan Donovan puts it,At every level of the tech stack, corporations are placed in positions to make value judgments regarding the legitimacy of content, including who should have access, and when and how.” And the decisions made by companies at varying layers of the stack are bound to have different impacts on free expression.

At the top of the stack are services like Facebook, Reddit, or Twitter, platforms whose decisions about who to serve (or what to allow) are comparatively visible, though still far too opaque to most users.  Their responses can be comparatively targeted to specific users and content and, most importantly, do not cut off as many alternatives. For instance, a discussion forum lies close to the top of the stack: if you are booted from such a platform, there are other venues in which you can exercise your speech. These are the sites and services that all users (both content creators and content consumers) interact with most directly. They are also the places people think of when they think of the content itself (i.e.,“I saw it on Facebook”). Users are often required to have individual accounts or advantaged if they do. Users may also specifically seek out the sites for their content. The closer to the user end, the more likely it is that sites will have more developed and apparent curatorial and editorial policies and practicestheir "signature styles." Finally, users typically have an avenue, flawed as it may be, to communicate directly with the service. 

At the other end of the stack are internet service providers (ISPs), like Comcast or AT&T. Decisions made by companies at this layer of the stack to remove content or users raise greater concerns for free expression, especially when there are few if any competitors. For example, it would be very concerning if the only broadband provider in your area cut you off because they didn’t like what you said onlineor what someone else whose name is on the account said. The adage “if you don’t like the rules, go elsewhere” doesn’t work when there is nowhere else to go.

In between are a wide array of intermediaries, such as upstream hosts like AWS, domain name registrars, certificate authorities (such as Let’s Encrypt), content delivery networks (CDNs), payment processors, and email services. EFF has a handy chart of some of those key links between speakers and their audience here. These intermediaries provide the infrastructure for speech and commerce, but many have only the most tangential relationship to their users. Faced with a complaint, takedown will be much easier and cheaper than a nuanced analysis of a given user’s speech, much less an analysis of the speech that might be hosted by a company that is a user of their services. So these service are more likely to simply cut a user or platform off than do a deeper review. At the same time, in many cases both speakers and audiences will not be aware of the identities of these support services and, even if they are, have no independent relationship with them. These services are thus not commonly associated with the speech that passes through them and have no "signature style" to enforce.

Infrastructure Takedowns Are Equally If Not More Likely to Silence Marginalized Voices

We saw a particularly egregious example of an infrastructure takedown just a few months ago, when Zoom made the decision to block a San Francisco State University online academic event featuring prominent activists from Black and South African liberation movements, the advocacy group Jewish Voice for Peace, and controversial figure Leila Khaled—inspiring Facebook and YouTube to follow suit. The decision, which Zoom justified on the basis of Khaled’s alleged ties to a U.S.-designated foreign terrorist organization, was apparently made following external pressure.

Although we have numerous concerns with the manner in which social media platforms like Facebook, YouTube, and Twitter make decisions about speech, we viewed Zoom’s decision differently. Companies like Facebook and YouTube, for good or ill, include content moderation as part of the service they provide. Since the beginning of the pandemic in particular, however, Zoom has been used around the world more like a phone company than a platform. And just as you don’t expect your phone company to start making decisions about who you can call, you don’t expect your conferencing service to start making decisions about who can join your meeting.

Just as you don’t expect your phone company to start making decisions about who you can call, you don’t expect your conferencing service to start making decisions about who can join your meeting.

It is precisely this reason that Amazon’s ad-hoc decision to cut off hosting to social media alternative Parler, in the face of public pressure, should be of concern to anyone worried about how decisions about speech are made in the long run. In some ways, the ejection of Parler is neither a novel, nor a surprising development. First, it is by no means the first instance of moderation at this level of the stack. Prior examples include Amazon denying service to WikiLeaks and the entire nation of Iran. Second, the domestic pressure on companies like Amazon to disentangle themselves from Parler was intense. After all, in the days leading up to its removal by Amazon, Parler played host to outrageously violent threats against elected politicians from its verified users, including lawyer L. Lin Wood.

But infrastructure takedowns nonetheless represent a significant departure from the expectations of most users. For one thing, they are cumulative, since all speech on the Internet relies upon multiple infrastructure hosts.  If users have to worry about satisfying not only their host’s terms and conditions but also those of every service in the chain from speaker to audience—even though the actual speaker may not even be aware of all of those services or where they draw the line between hateful and non-hateful speech—many users will simply avoid sharing controversial opinions altogether. They are also less precise. In the past, we’ve seen entire large websites darkened by upstream hosts because of a complaint about a single document posted. More broadly, infrastructure level takedowns move us further toward a thoroughly locked-down, highly monitored web, from which a speaker can be effectively ejected at any time.

Going forward, we are likely to see more cases that look like Zoom’s censorship of an academic panel than we are Amazon cutting off another Parler. Nevertheless, Amazon’s decision highlights core questions of our time: Who should decide what is acceptable speech, and to what degree should companies at the infrastructure layer play a role in censorship?

At EFF, we think the answer is both simple and challenging: wherever possible, users should decide for themselves, and companies at the infrastructure layer should stay well out of it. The firmest, most consistent, approach infrastructure chokepoints can take is to simply refuse to be chokepoints at all. They should act to defend their role as a conduit, rather than a publisher. Just as law and custom developed a norm that we might sue a publisher for defamation, but not the owner of the building the publisher occupies, we are slowly developing norms about responsibility for content online. Companies like Zoom and Amazon have an opportunity to shape those norms—for the better or for the worse.

Internet Policy and Practice Should Be User-Driven, Not Crisis-Driven

It’s easy to say today, in a moment of crisis, that a service like Parler should be shunned. After all, people are using it to organize attacks on the U.S. Capitol and on Congressional leaders, with an expressed goal to undermine the democratic process. But when the crisis has passed, pressure on basic infrastructure, as a tactic, will be re-used, inevitably, against unjustly marginalized speakers and forums. This is not a slippery slope, nor a tentative prediction—we have already seen this happen to groups and communities that have far less power and resources than the President of the United States and the backers of his cause. And this facility for broad censorship will not be lost on foreign governments who wish to silence legitimate dissent either. Now that the world has been reminded that infrastructure can be commandeered to make decisions to control speech, calls for it will increase, and principled objections may fall to the wayside. 

Over the coming weeks, we can expect to see more decisions like these from companies at all layers of the stack. Just today, Facebook removed members of the Ugandan government in advance of Tuesday’s elections in the country, out of concerns for election manipulation. Some of the decisions that these companies make may be well-researched, while others will undoubtedly come as the result of external pressure and at the expense of marginalized groups.

The core problem remains: regardless of whether we agree with an individual decision, these decisions overall have not and will not be made democratically and in line with the requirements of transparency and due process. Instead they are made by a handful of individuals, in a handful of companies, the most distanced and least visible to the most Internet users. Whether you agree with those decisions or not, you will not be a part of them, nor be privy to their considerations. And unless we dismantle the increasingly centralized chokepoints in our global digital infrastructure, we can anticipate an escalating political battle between political factions and nation states to seize control of their powers.