Shining a Light on Censorship: How Transparency Can Curtail Government Social Media Censorship and More

Endeavoring to address high‐​profile instances of censorship by proxy during the COVID-19 pandemic and 2020 election, the U.S. House of Representatives recently passed the Protecting Speech from Government Interference Act.³ If enacted, this legislation would broadly prohibit federal employees from censoring with respect to social media.⁴ The problem with this and other prohibition‐​based approaches to censorship by proxy is that it is practically impossible to precisely define what conduct should be prohibited.

In matters of national security, law enforcement, and beyond, government officials regularly make statements that encourage private actors to suppress information, and not all of this is objectionable. Consider, for example, an FBI agent who requests that a newspaper delay publishing certain details about an ongoing criminal investigation because doing so could undermine attempts to capture the suspect. Any practically workable prohibition will be far too narrow, and any prohibition broad enough to cover the field will inevitably prove unworkable when it sweeps in routine government activities.

There is an easier and more effective way to address censorship by proxy: transparency. Federal officials should be required to publicly report attempts to suppress Americans’ exercise of speech and associational rights. Censorship by proxy, as practiced today, depends on secrecy and practical obscurity to evade public and legal accountability. Forcing attempted censorship out of the shadows stands to deter the worst abuses and ensure that officials who aren’t deterred can be held to account. At the same time, a transparency‐​based approach avoids the difficulties inherent to prohibition because it dispenses with the need to precisely define the prohibited conduct; in this context, over‐​inclusiveness in disclosure is no vice.

Ultimately, a transparency‐​based approach to censorship by proxy would be more workable than prohibition and just as effective in deterring improper efforts by government officials to suppress speech. “Sunlight,” after all, “is said to be the best of disinfectants.”⁵

Jawboning is not a new phenomenon. The term was coined in the 1970s to describe World War II–era activities of the Office of Price Administration and Civilian Supply, although government officials assuredly engaged in jawboning prior to that time.⁶ While censorship by proxy is not new, as the internet and social media have grown in importance the government has increasingly “sought to enlist private actors within the chain as proxy censors to control the flow of information.”⁷ Congressional jawboning is by now a well‐​documented problem, albeit one that more often involves hectoring than any realistic threat of legislative action.⁸ Executive branch jawboning, by contrast, gains force from the administrative state’s vast regulatory and enforcement authority and practically unchecked discretion in wielding it. The threat of regulatory action, whether implicit or overt, presses private actors to comply with government requests and suggestions no less than directives would.

For example, during the COVID-19 pandemic, the White House press secretary admitted that federal officials were flagging for Facebook “problematic” posts that spread “disinformation.”⁹ Following White House threats, Twitter suspended the account of a former New York Times reporter, Alex Berenson, for criticizing the COVID-19 vaccines.¹⁰ And, beginning during the 2020 election, FBI officials encouraged social media companies to be cautious about misinformation and foreign interference.¹¹ Before the election, the FBI held regular meetings about election misinformation with a broad range of tech companies, “including Twitter, Facebook, Reddit, Discord, Wikipedia, Microsoft, LinkedIn, and Verizon Media.”¹²

Censorship by proxy presents a serious risk to free speech in the United States. The First Amendment severely limits the government’s ability to directly censor speech. But “by working through intermediaries, government can suppress speech quickly, without broad support, and potentially without alerting anyone of its involvement.”¹³ The practice serves to end‐​run the First Amendment.

Before considering proposals to address censorship by proxy, it is important to understand a few things. First, and most significantly, government officials appear to rarely resort to outright demands to press third parties into censoring. As the Twitter Files revealed, most censorship by proxy occurs in the form of requests or suggestions. In light of the government’s pervasive regulatory power over social media companies, banks, and other institutions, mere requests for censorship or questions about controversial content are often perceived as instructions, if not veiled threats. At a minimum, service providers want to stay on the good side of their regulators, and so they will almost always go along with what they understand the regulators to want.¹⁴ Any approach to censorship by proxy that addresses only demands or overt coercion will be woefully underinclusive.

Second, it is conceptually important to separate government‐​directed censorship by proxy from decisions by private parties, including social media companies, to suppress speech, if those decisions are made without government interference. While the independent ability of social media companies to suppress speech on their platforms may raise its own concerns, attempting to regulate private actors’ decisions raises constitutional and policy issues that are beyond the scope of this analysis. Whatever one thinks of “social media censorship,” government‐​directed censorship by proxy violates fundamental free speech principles—and potentially the First Amendment—and is worth addressing in its own right.

Third, censorship by proxy is not limited to social media. The government targets many types of service providers—from banks to law firms—and pressures those providers to deny services to individuals or entities with disfavored views, thus leading to censorship. For example, Maria Vullo, the former superintendent of the New York State Department of Financial Services, pressured banks and insurance companies to stop doing business with the National Rifle Association following a shooting in Parkland, Florida.¹⁵ The Obama administration’s Operation Chokepoint similarly pressured banks to stop doing business with gun and ammunition dealers, payday lenders, and other disfavored businesses.¹⁶ And, after the House of Representatives hired attorneys at BakerHostetler to represent the House in a lawsuit against the Obama administration, Health and Human Services officials pressured the general counsels of health care companies to stop doing business with the law firm.¹⁷ Any policy approach that is limited to social media will miss areas where government officials are able to wield significant power and influence to suppress constitutionally protected speech and association.

Finally, not all government requests are constitutionally objectionable. Law enforcement officials may ask social media platforms to suppress illegal speech such as child sexual abuse material, as well as information that threatens to undermine the integrity of an ongoing criminal investigation. CIA officials may request the removal of legitimately classified national security information. And should the government really be prohibited from discussing foreign‐​influence campaigns with social media companies? At least some of the government’s reports of foreign disinformation have proven to be incorrect, but it is not difficult to imagine that social media companies may legitimately want to know that foreign actors are abusing their services so that they can choose how to respond.

Critics of censorship by proxy have identified two primary ways to address it: First Amendment litigation and proposed legislation banning censorship by proxy. Neither approach, however, offers a fully satisfactory solution.

The First Amendment lawsuits are rooted in Bantam Books, Inc. v. Sullivan, where the Supreme Court held that the actions of the Rhode Island Commission to Encourage Morality in Youth violated the First Amendment.¹⁸ The commission sent notices to distributors of books “that certain designated books or magazines … had been declared by a majority of its members to be objectionable for sale distribution or display” to minors.¹⁹ The notice thanked the distributors in advance for their cooperation, and reminded them of its “duty to recommend to the Attorney General prosecution of purveyors of obscenity.”²⁰ The commission also sent lists of what it considered to be objectional publications to local police departments, and local police officers would typically visit a distributor after “receipt of a notice to learn what action he had taken.”²¹

Notwithstanding the fact that the commission merely “exhort[ed] booksellers and advise[d] them of their legal rights” and did not directly censor books, the Court held that its “operation was in fact a scheme of state censorship effectuated by extralegal sanctions.”²² The Court explained that “the threat of invoking legal sanctions and other means of coercion, persuasion, and intimidation” used by the commission “amply demonstrates that the Commission deliberately set out to achieve the suppression of publications deemed ‘objectionable’ and succeeded in its aim” in violation of the First Amendment.²³

Litigants have relied on Bantam and related precedent to challenge modern censorship by proxy with mixed success. For example, former president Trump filed a class action lawsuit against Twitter and the United States, alleging that the government coerced Twitter into suppressing speech in violation of the First Amendment; the complaint was dismissed on the grounds that Twitter was a private actor, but Trump has appealed.²⁴ In Back​page​.com, LLC v. Dart, the Seventh Circuit held that a county sheriff violated the First Amendment by crossing the line between “attempts to convince and attempts to coerce.”²⁵ The sheriff had sent letters to Visa and MasterCard requesting them to “immediately cease and desist from allowing your credit cards to be used to place ads on websites like Back​page​.com”; intimated that the companies “could be prosecuted for processing payments made by purchasers of the ads on Backpage that promote unlawful sexual activity”; and requested “contact information for an individual within” their company that the sheriff could “work with … on this issue.”²⁶

Similarly, in Missouri v. Biden, a federal district court held that the complaint stated a claim for a violation of the First Amendment based on allegations that Biden administration officials had improperly threatened social media companies to censor posts on social media that spread “disinformation” about the pandemic and other topics.²⁷ The complaint alleged numerous “threats, some thinly veiled and some blatant, made by Defendants in an attempt to effectuate its censorship program.”²⁸

Subsequently, the district court entered a far‐​reaching injunction barring numerous government officials and agencies from “urging, encouraging, pressuring, or inducing in any manner the removal, deletion, suppression, or reduction of content containing protected free speech posted on social‐​media platforms.”²⁹ At the same time, however, the injunction specifically excludes many actions that may amount to censorship by proxy, such as “exercising permissible public government speech promoting government policies or views on matters of public concern”; “informing social‐​media companies of threats that threaten the public safety or security of the United States”; and countering “foreign attempts to influence elections.”³⁰ While these exceptions track the government’s well‐​established power to speak and its authority in such fields as national security, they also illustrate how even broad First Amendment remedies may fail to reach many of the most controversial government communications.

For example, the Missouri v. Biden plaintiffs allege that the FBI pressed social media companies to suppress the Hunter Biden laptop story by insinuating that it was Russian disinformation or by failing to correct the misimpression that it was—things that would seem to be exempted from the injunction.³¹ On appeal, the Fifth Circuit affirmed the district court in part but vacated certain prohibitions of the preliminary injunction—such as its application to speech “urging, encouraging, pressuring, or inducing” content moderation that did not rise to the level of “coercion or significant encouragement”—as overbroad and vague, making the injunction even less effective [emphasis added].³²

In National Rifle Association v. Vullo, the Second Circuit held that Vullo’s letter to banks and insurance companies warning them of “reputational risks … that may arise from their dealings with the NRA or similar gun promotion organizations” did not “cross the line between an attempt to convince and an attempt to coerce.³³ Although Vullo “plainly favored gun control over gun promotion and sought to convince DFS‐​regulated entities to sever business relationship with gun promotion groups,” her statements were “clear examples of permissible government speech.” They had an “even‐​handed, nonthreatening tone and employed words intended to persuade rather than intimidate.”³⁴

The Ninth Circuit similarly rejected the claim of a Twitter user whose tweets had been flagged as misleading in O’Handley v. Weber.³⁵ Twitter flagged the tweets after California’s Office of Elections Cybersecurity—whose mission is to “monitor and counteract false or misleading information regarding the electoral process”—had submitted a report through Twitter’s Partner Support Portal, which allows government agencies to initiate an expedited review process for tweets they believe violate Twitter’s terms of service.³⁶ The Ninth Circuit rejected the claim because Twitter “acted under the terms of its own rules” in flagging the tweet, and the government did not “threaten adverse action to coerce a private party into performing a particular act.”³⁷

These precedents demonstrate that it will be difficult for First Amendment litigation to operate as a comprehensive check on censorship by proxy. Courts have generally found only the most egregious government conduct to be sufficiently coercive to violate the First Amendment. But even when government officials speak in an even‐​handed voice, their implied threats and encouragement can cause third parties to suppress speech. The practical result of such conduct can be indistinguishable from direct government censorship or coercive threats. Accordingly, there is little chance that the prospect of First Amendment litigation will deter government officials from censoring by proxy.

Broader problems exist with relying on First Amendment litigation to check abusive government censorship. The individuals who know that they were the targets of government censorship—and therefore that their First Amendment rights were potentially violated—constitute a small fraction of the victims of censorship by proxy. And even those individuals may not know enough facts about the government’s targeting to draft a complaint with allegations sufficient to survive a motion to dismiss under the various legal standards that the courts have applied to censorship by proxy claims. Litigation also consumes significant time and resources, making it an even poorer potential tool to address jawboning at scale.

In light of the limitations of existing law, some policymakers have advocated for legislation prohibiting censorship by proxy. But legislation that bans jawboning needs to define it perfectly, lest an overbroad definition sweep up routine government activities or otherwise prove to be unworkable. The most prominent prohibition‐​based proposal—the Protecting Speech from Government Interference Act, which was passed by the House in March 2023—illustrates this problem.³⁸ The bill would prohibit federal employees from, among other things, using their authority to censor private entities, which it would define to mean “influencing or coercing, or directing another to influence or coerce, for … the removal or suppression of lawful speech, in whole or in part, from or on any interactive computer service,” as well as the addition of disclaimers on or the removal of a user for such speech. The bill contains a limited exception for law enforcement, providing that it should not “be construed to prohibit an employee from engaging in lawful actions against unlawful speech within the official authority of such employee for the purpose of exercising legitimate law enforcement functions.”³⁹

Among other problems, the bill’s definition of the sort of actions that constitute prohibited censorship by proxy—“influencing or coercing”—is overbroad. What does it mean for a government official to influence a private entity to suppress speech? The term “influence” is understandably added to “coerce” in the bill in order to cover actions similar to those in Weber and Vullo. However, all sorts of routine government actions could influence a private entity to suppress speech, such as a statement in a press conference by a Food and Drug Administration (FDA) official that ivermectin is not approved to treat COVID-19. That statement might ultimately lead Twitter and Facebook to add disclaimers to certain posts, but that is not a good justification for suppressing the official’s speech, which is important to inform the public about the FDA’s views.

The bill also would prohibit routine and arguably beneficial government speech, such as requests made to social media for law enforcement or national security reasons. Although the bill contains an exception for law enforcement, it only permits law enforcement to suggest the suppression of unlawful speech. This exception is far too narrow. Consider the example of a law enforcement officer who requests that a reporter delay publishing certain details about a crime because publicizing those details could undermine an ongoing criminal investigation. There is nothing unlawful about the reporter publishing those details, and thus the bill would prohibit the officer’s request.

It would be possible to craft a narrower bill—such as one limited to prohibiting official coercion, with broader exceptions for law enforcement and other routine government statements. While such legislation would be more workable in practice, it would be too narrow to cover most of the troubling examples of censorship that do not involve overt coercion—the same shortcoming seen in the Missouri v. Biden injunction.

Moreover, neither litigation nor the prohibition‐​based approach offers a solution to one of the most pressing problems presented by censorship by proxy: most individuals targeted by such censorship never know about it. When Twitter or Facebook suppress posts at the government’s behest, the users only know that their posts have been deleted but not of the government’s involvement. “Neither the private intermediary nor the government officials will ordinarily have much motivation to acknowledge when jawboning occurs. People whose speech has been suppressed will therefore not know that they can challenge that suppression on constitutional grounds.”⁴⁰ To address this problem, there are some legislative transparency proposals that would require social media platforms to publish year‐​end reports relating to their content moderation. But reports under such proposals would “offer little insight into particular content moderation decisions,” and requiring platforms to disclose more information about individual decisions would likely be impractical and may “violate First Amendment protections of editorial privilege.”⁴¹

There is an easier way to address censorship by proxy: require government officials to disclose it. We know that transparency and accountability work in this space. The reason government officials carry out these activities in the shadows is that jawboning doesn’t stand up to the light of day. When word leaked about the Department of Homeland Security’s proposed Disinformation Governance Board, the result was public outrage and the board was scrapped almost overnight.⁴² The Twitter Files shed light on how government officials interact with social media networks and led to congressional hearings and a new House subcommittee on the Weaponization of the Federal Government.⁴³

Disclosure would also facilitate litigation by those whose rights have been violated by providing the crucial missing link between a government request and an adverse action, such as termination of a social media account. Litigants in jawboning cases, such as Trump v. Twitter and AAPS v. Schiff, have struggled to establish a link between the removal of their speech and a particular government request.⁴⁴ Such evidence has often only been made available after a lengthy and expensive discovery process—which itself is only available to plaintiffs who have enough facts to plausibly allege government action. Transparency would reveal the hand of government, where it exists, from the get‐​go.

This is why Meta’s oversight board has endorsed transparency, proposing that Facebook “report regularly on state actor requests to review content.”⁴⁵ For service providers, disclosure would remove the shadow of potential government inference that hangs over their content moderation, while still allowing platforms to receive government reports about bad actors.⁴⁶

Importantly, while it is difficult or impossible to draw a clear line separating permissible persuasion from blameworthy censorship by proxy, a transparency‐​based approach doesn’t have to cut so fine. It can be overbroad because nothing is being prohibited. It doesn’t take a position on whether any particular act is right or wrong, only that the public and the people being targeted should know what the government is doing. We outline below, in broad strokes, the key features of legislation adopting a transparency‐​based approach.

The first component of a transparency‐​based approach would be to require all federal employees to report to the Office of Management and Budget (OMB) any request to a private service provider to limit or deny services based on activities protected by the First Amendment. The report would include information about the federal employee who made the request, the date of the request, and certain details about the request, such as the reason it was made and the speech that was targeted.

A few features of the reporting requirement bear noting. First, it would require reporting of requests beyond the “coercive” acts limited by the First Amendment. By including circumstances where a government official “suggests” or “encourages” a third party to suppress speech or association, the legislation would require reporting of the sorts of censorship seen in Weber and Vullo. Second, the requirement would apply to acts that target any service provider, such as a bank or credit card company, and not just social media companies. Given the large number of third‐​party entities that can be enlisted by the government in its effort to suppress speech and association, proposals that are limited to social media companies are far too narrow.

Finally, it is important to note that there is ample precedent for this requirement. The Paperwork Reduction Act, for example, requires federal agencies to report proposed collections of information to the OMB and publish notices of their collection activities in the Federal Register.⁴⁷ And covered federal employees report to the government, through simple online portals, contacts with the media and unofficial foreign travel.⁴⁸ The reporting requirement need not be burdensome—a report could be submitted in just a couple minutes.

The OMB would be responsible for assembling the reports and preparing them for prompt public disclosure on a centrally administered website. In order to protect privacy, national security, and law enforcement efforts, the public reports should be subject to certain redactions. Fortunately, Congress has already balanced the interest in government transparency against these concerns in the exemptions and exclusions to the Freedom of Information Act, which requires agencies to publicly disclose certain information.⁴⁹

Among other things, the act exempts from disclosure classified information, trade secrets, certain law enforcement information, and material “the disclosure of which would constitute a clearly unwarranted invasion of personal privacy.”⁵⁰ Transparency‐​based censorship by proxy legislation should incorporate these preexisting exemptions and exclusions and also require the redaction of any targeted individual’s personally identifying information.

The need to redact these categories of information from the public reports is largely self‐​explanatory. If a government official flagged a post on Twitter that disclosed sensitive classified information, for example, it would be counterproductive to require the OMB to publicly disclose the classified information. Similarly, if the government publicly disclosed the names of individuals whose speech was targeted, that would intrude upon the privacy of those individuals and potentially subject them to threats or other social sanctions—certainly not the intended result of legislation aimed at protecting their First Amendment rights. The OMB should therefore redact such information from its public reports, and transparency‐​based legislation should make clear that the OMB is not required to publicly report it.

Because the public reports generally would not include personally identifying information, transparency‐​based legislation should also provide for notification, when feasible, to any individual whose speech was targeted. This could be easily accomplished in most cases by requiring that the service provider notify the user or customer. Notification would enable the individual to decide whether to go public and identify themselves, and to decide whether to pursue any claim that the jawboning violates the First Amendment.

The notification would also be subject to redactions, although the interest in confidentiality differs from the public disclosure context. For example, it would not invade the privacy of the individual receiving the notification to have their personally identifying details included in the notification. We therefore suggest incorporating the Privacy Act’s exemptions for purposes of the individual notification requirement because its exemptions strike an appropriate balance between individual disclosure and the government’s interest in confidentiality.⁵¹

Requiring intermediaries to notify users of government requests is the best practical way of preserving user anonymity. Having the government inform users directly would require the government knowing who they are and how to contact them. So long as notification requirements are kept simple, such as by using whatever contact information platforms already have on file or by issuing notifications within the platform itself, notification will be minimally burdensome for platforms and minimally invasive for users.

Among other enforcement mechanisms, the legislation should include penalties for federal employees who willfully fail to make a complete and accurate report of their requests to service providers. These penalties could mirror those in the Hatch Act, which prohibits federal employees from engaging in certain political activities and subjects violators to disciplinary consequences, such as termination and civil fines.⁵² The threat of these or other penalties would deter federal employees from failing to report their jawboning to the OMB.

We suggest a six‐​year statute of limitations for enforcement actions. A statute of limitations of this length will mean that federal employees cannot willfully fail to disclose censorship by proxy under the assumption that a currently favorable administration will decline to enforce the statute against them. Efforts to evade accountability by hiding or deleting past requests would violate the Federal Records Act and may be subject to criminal prosecution.⁵³

Notably, while the foregoing analysis assumes that a transparency‐​based approach would be adopted by legislation, almost all of these elements could be implemented through executive action. The president has broad inherent and statutory authority to “prescribe regulations for the conduct of employees in the executive branch.”⁵⁴ Establishing rules governing censorship by proxy, including transparency measures, falls squarely within this authority.⁵⁵

If a transparency‐​based approach were adopted by executive action, there would be no specific penalties for federal employees who violate its provisions. If a federal employee failed to disclose a request to a service provider, however, they would be disobeying a lawful job requirement and therefore be subject to disciplinary measures. Moreover, executive branch regulations also could require employees to certify on a regular basis that they have disclosed all reportable requests. For employees who willfully failed to do so, this certification would be a materially false statement to the government and the employee would be subject to criminal fines and prosecution.⁵⁶

Policymakers are rightly concerned with censorship by proxy, which permits government officials to evade the First Amendment by encouraging or threatening service providers to censor disfavored speech. While existing law is inadequate to address this problem, proposals that would prohibit all government communication with service providers are ill‐​advised. A better and easier solution is to require government officials to disclose their requests, subjecting them to public scrutiny and ultimately deterring government speech suppression.

The views expressed in this paper are those of the authors and do not necessarily reflect the views of any entities with whom they are affiliated or represent.