Daphne Keller
In recent
years, lawmakers around the world have proposed a lot of new intermediary
liability (IL) laws. Many have been miscalibrated – risking serious collateral damage without necessarily using the best
means to advance lawmakers’ goals. That shouldn’t be a surprise. IL isn’t like
tax law or farm subsidies. Lawmakers, particularly in the United States, haven’t
thought much about IL in decades. They have little institutional knowledge
about which legal dials and knobs can be adjusted, and what consequences to
expect.
This post will lay
out a brief menu, framed for a U.S. audience, of IL legal mechanisms. Most are
relatively well-understood from laws and literature around
the world; a few are
newly emerging ideas. It foregrounds legislative choices that affect free
expression, but does not try to identify hard limits created by the First
Amendment or other free expression laws.
Of course,
crafting laws isn’t really like ordering off a menu. It’s more like cooking:
the ingredients intermingle and affect one another. A law holding platforms
liable for defamatory speech they “know” about, for example, may mean something
different depending whether the law lets accused speakers explain and defend
their posts. But isolating the options in modular form can, I hope, help in
identifying options for pragmatic and well-tailored laws.
IL laws generally
try to balance three goals. The first is preventing harm. It’s no accident that
intermediary immunities are typically weakest for content that poses the greatest
threats, including material criminalized by U.S. federal law. The second is
protecting speech and public participation. For this goal, one concern is to avoid
over-removal – the well-documented phenomenon of platforms cautiously deferring
to bogus legal accusations and taking down users’ lawful speech. Another is to encourage
new market entrants to build, and investors to fund, open speech platforms in
the first place. The third, related goal is encouraging technical innovation
and economic growth. A rule that creates great legal uncertainty, or that can
only be enforced by hiring armies of moderators, raises formidable barriers to
entry for potential competitors with today’s mega-platforms. Lawmakers use the
doctrinal dials and knobs listed in the remainder of this post to adjust policy
trade-offs between these goals.
Major Free Expression Considerations
Who decides what speech is illegal?
Outside the United States, blanket immunities like CDA 230 are rare. But it’s not uncommon for courts or legislatures to keep platforms out of the business of deciding what speech violates the law. One IL model widely endorsed by free expression advocates holds platforms immune unless a court or other government authority rules content illegal. In practice, this highly speech-protective standard typically has exceptions, requiring platforms to act of their own volition against highly recognizable and dangerous content such as child sex abuse images. Lawmakers who want to move the dial more toward harm prevention without having platforms adjudicate questions of speech law can also create accelerated administrative or TRO processes, or give platforms other responsibilities such as educating users, developing streamlined tools, or providing information to authorities.
Must platforms proactively monitor,
filter, or police users’ speech?
Human rights literature includes strong warnings against making platforms monitor their users. Many IL laws expressly bar such requirements, though they have gained traction in recent European legislation. One concern is that technical filters are likely to over-remove, given their inability to recognize contexts like news reporting or parody. (However, filtering is relatively accepted for child sexual abuse images, which are unlawful in every context.) Another is that, when platforms have to review and face over-removal incentives for every word users post, the volume and invasiveness of unnecessary takedowns can be expected to rise. Legal exposure and enforcement costs under this model may also give platforms reason to allow only approved, pre-screened speakers – and deter new market entrants from challenging incumbents.
Must platforms provide “private due
process” in takedown operations?
Improving platforms’ internal notice-and-takedown processes can protect against over-removal. A widely supported civil society document, the Manila Principles, provides a list of procedural rules for this purpose. For example, a platform can be required or incentivized to notify speakers and let them defend their speech – which may help deter bad-faith notices in the first place. Accusers can also be required to include adequate information in notices, and face penalties for bad-faith takedown demands. And platforms can be required to disclose raw or aggregate data about takedowns, in order to facilitate public review and correction.
Can platforms’ use of private Terms of
Service prohibit lawful expression?
Platforms often prohibit disfavored but legal speech under their Terms of Service (TOS). To maximize users’ free expression rights, a law might limit or ban this restriction on speech. In the United States, though, such a law might violate platforms’ own speech and property rights. Platforms’ value for ordinary users would also decline if users were constantly faced with bullying, racial epithets, pornography, and other legal but offensive matter. (I address relevant law in depth here and explore possible regulatory models in that paper’s final section.)
Can speakers defend their rights in
court?
Platform over-removal incentives come in part from asymmetry between the legal rights of accusers and those of speakers. Victims of speech-based harms can often sue platforms to get content taken down. Speakers can almost never sue to get content reinstated. A few untested new laws in Europe try to remedy this, but it is unclear how well they will work or how speakers’ claims will intersect with platforms’ power to take down speech using their TOS.
Are leaving content up and taking it
down the only options?
IL laws occasionally use more tailored remedies, in place of binary take-down/leave-up requirements – like making search engines suppress results for some search queries, but not others. Platforms could also do things like showing users a warning before displaying certain content, or cutting off ad revenue or eligibility for inclusion in recommendations. In principle, IL law could also regulate the algorithms platforms use to rank, recommend, or otherwise amplify or suppress user content – thought that would raise particularly thorny First Amendment questions and be extremely complex to administer.
Treating Platforms Like
Publishers
Making platforms liable for content they
control
Most IL laws strip immunity from platforms that are too actively involved in user content. Some version of this rule is necessary to distinguish platforms from content creators. More broadly, putting liability on an entity that exercises editor-like power comports with traditional tort rules and most people’s sense of fairness. But standards like these may play out very differently for Internet platforms than for traditional publishers and distributors, given the comparatively vast amount of speech platforms handle and their weak incentives to defend it. Laws that reward passivity may also deter platforms from trying to weed out illegal content and generate legal uncertainty about features beyond bare-bones hosting and transmission.
Making platforms liable for content they
know about
Many legal systems hold platforms liable for continuing to host or transmit illegal content once they “know” or “should know” about it. Laws that rely on these scienter standards can protect legal speech somewhat by defining “knowledge” narrowly or adding elements like private due process. Other legal regimes reject scienter standards, considering them too likely to incentivize over-removal.
Using “Good Samaritan” rules to
encourage content moderation
Platforms may be deterred from moderating content by fear that their efforts will be used against them. Plaintiffs can (and do) argue that by moderating, platforms assume editorial control or gain culpable knowledge. Concern about the resulting perverse incentives led Congress to create CDA 230, which makes knowledge and control largely irrelevant for platform liability. This encouraged today’s moderation efforts but also introduced opportunities for bias or unfairness.
Different Rules for
Different Problems
Legal claims
IL laws often tailor platforms’ duties based on the claim at issue. For example, they may require urgent responses for particularly harmful content, like child sex abuse images; deem court review essential for claims that turn on disputed facts and nuanced law, like defamation; or establish private notice-and-takedown processes in high-volume areas, like copyright.
Platform technical function
Many IL laws put the risk of liability on the entities most capable of carrying out targeted removals. Thus, infrastructure providers like ISPs or domain registries generally have stronger legal immunities than consumer-facing platforms like YouTube, which can do things like take down a single comment or video instead of a whole page or website.
Platform size
Recently, experts have raised the possibility of special obligations for mega-platforms like Google or Facebook. Drafting such provisions without distorting market incentives or punishing non-commercial platforms like Wikipedia would be challenging. In principle, though, it might improve protections on the most popular forums for online expression, without imposing such onerous requirements that smaller market entrants couldn’t compete.
General Regulatory Approach
Bright-line rules versus fuzzy standards
IL rules can hold platforms to flexible standards like “reasonableness,” or they can prescribe specific steps. Platforms – especially the ones that can’t hire a lawyer for every incoming claim – typically favor the latter, because it provides relative certainty and guidance. Free expression advocates also often prefer clear processes, because they reduce the role of platform judgment and allow legislatures to add procedural protections like counter-notice.
Liability for single failures versus
liability for systemic failures
Some recent European laws and proposals accept that takedown errors are inevitable and do not impose serious financial penalties for individual items of content. Instead they penalize platforms if their overall takedown system is deemed inadequate. This approach generally reduces over-removal incentives, but is more viable in legal systems with trusted regulators.
Liability for platforms versus liability
for speakers
Internet users may see little reason to avoid disseminating unlawful content when the legal consequences of their actions fall primarily on platforms. Laws could be structured to shift more risk to those individuals. For example, claims against platforms could be limited if claimants do not first seek relief from the responsible user. Or platforms’ immunities could be made contingent on preserving or disclosing information about online speakers – though this would raise serious concerns about privacy and anonymity rights.
Daphne Keller
is Director of Intermediary Liability at the Stanford Center for Internet and
Society, and was previously Associate General Counsel at Google. She can
be reached at daphnek@law.stanford.edu.