Civics · Law · Geopolitics: The Boundaries of Speech & Information

First ask: why protect speech at all? The mainstream answers come in three layers—discovering truth (let ideas compete so error can be falsified: the "marketplace of ideas"), democratic self-government (citizens must be able to criticize power to hold it accountable), and individual autonomy (expression is part of personhood). Precisely because the reasons to protect it run so deep, the bar to restrict it is set deliberately high.

Yet no system makes it absolute. The real disagreement is where the line falls: nearly every jurisdiction declines to protect direct incitement to imminent violence, true threats, and defamation; the disputes cluster in gray zones—"hate speech," "offense," "denial of history"—tolerated as opinion, or banned as harm.

• United States: a very high bar. The 1969 Brandenburg v. Ohio established the "imminent lawless action" test—speech can be banned only when it is intended and likely to incite imminent lawless action, replacing the looser earlier "clear and present danger." The result: even extreme, offensive speech is mostly protected.
• Germany: militant democracy. Drawing on its Nazi past, Germany criminalizes Holocaust denial and the display of Nazi symbols, writing into law the idea of not letting enemies of freedom use freedom to destroy it (streitbare Demokratie).

Core tension: the harm principle vs. the dignity principle. The American view (steelman): unless speech causes immediate, concrete harm, government stays out entirely—because "who decides what is harmful" is itself dangerous, and the moment it is opened it gets abused to silence dissent; better to tolerate the offensive. The continental view (steelman): some speech (racial hatred, denial of genocide) attacks the equal dignity of specific groups and poisons public life, and standing aside makes the vulnerable bear the cost. Not right vs. wrong, but which way to tilt the margin of error.

Misconception: "The law has long said you can't shout 'fire' in a theater." This line has been misused for a century. It comes from Justice Holmes's analogy (dictum) in the 1919 Schenck case—never a binding rule; and the original said "falsely shouting." More importantly, that lenient standard was replaced in 1969 by Brandenburg. Citing this phrase to prove "of course speech can be freely restricted" invokes something that stopped being the law long ago.

"Censorship" is not one thing; mechanically it splits into two, with sharply different costs. Prior restraint: speech is stopped before it is uttered—publishing needs approval, content needs pre-clearance, sites need a license. Post-hoc punishment: speech may go out first, but if it breaks the law, responsibility is imposed afterward (damages, fines, conviction).

Why do legal traditions generally fear prior restraint most? Two hidden costs: first, the blocked speech is never seen by the public, so there is no chance even to debate whether it should have been blocked; second, it produces a chilling effect—to pass review, people pre-emptively censor out large amounts of otherwise lawful expression. Post-hoc punishment at least lets speech enter the public sphere and be tested first.

• Strong guard against prior restraint: the US constitutional tradition sets a near-insurmountable bar against it—in the 1971 "Pentagon Papers" case, the government invoked national security to block newspapers from publishing, yet the Supreme Court refused the prior injunction, preferring to weigh liability afterward.
• The licensing spectrum: many countries keep licensing and registration regimes for press and broadcast; a mild version handles technical matters like spectrum allocation, while a strict version becomes a gate that screens who may speak at all.
• New pre-gates in the digital age: today the real "prior gate" often sits in platform upload filters and algorithmic ranking—content is auto-blocked or down-ranked before anyone sees it, migrating the classic "government prior censorship" problem onto private infrastructure.

The before-vs-after trade-off. For some prior control (steelman): certain harms (leaked military secrets, child sexual abuse material, imminent incitement to violence) can't be undone once out; punishing afterward is too late, so they must be stopped at the source. Against prior restraint (steelman): it hands the "lawful or not" judgment to an invisible, hard-to-hold-accountable gate that tends to "block when in doubt," systematically shrinking the space for lawful expression; except in a very narrow set of cases, use post-hoc accountability instead. Most rule-of-law states compromise: prior restraint confined to a narrow list, the rest left to after-the-fact liability.

Misconception: "Only governments can censor." At the constitutional level, most countries' free-speech clauses indeed constrain government—a private platform deleting your post usually isn't "censorship" in the legal-constitutional sense. But in practical effect, when a few platforms control the main gateways to public debate, their content policies may shape "who gets heard" no less than a government does—which is exactly why "platform liability" is such a hard problem (see next section).

The core problem is called intermediary liability: for unlawful content users post (defamation, hate, fraud), should the platform answer for every item like a "publisher," or merely carry it like a "phone company"? Set liability too heavy and platforms, to protect themselves, over-delete (sweeping away lawful content too); set it too light and harmful content goes unaddressed. Countries offer three typical solutions:

• Broad immunity: platforms are in principle not liable for user content, and don't become "publishers" merely by removing some of it.
• Notice-and-takedown + deadlines: once notified of, or facing obviously unlawful content, they must remove it within a time limit or face heavy fines.
• Duty of care: platforms must build a whole apparatus of risk assessment, transparent appeals, and algorithmic disclosure—proving "due diligence."

• United States: Section 230. The 1996 Communications Decency Act's Section 230—called "the 26 words that created the internet"—provides that platforms shall not be treated as the publisher of user content, so they are neither liable for it nor lose immunity by moderating it. It is seen as the legal cornerstone of the US platform economy—and a lightning rod for controversy.
• Germany: NetzDG. The 2017 Network Enforcement Act requires large platforms to remove "manifestly unlawful" content within 24 hours and other unlawful content within 7 days, with fines up to €50 million—pushing liability onto platforms, but also criticized for creating incentives to "over-remove."
• EU: the DSA. The 2022 Digital Services Act uses a tiered approach to harmonize national rules: baseline duties for all intermediaries, and the heaviest risk-assessment, algorithmic-transparency, and external-audit duties for "Very Large Online Platforms" (over 45 million EU monthly users)—shifting from "how fast you delete" to "how sound your governance process is."

The immunity-vs-accountability dilemma. For immunity (steelman): if platforms must answer for every user post, the rational move is to delete when in doubt, wiping out vast amounts of lawful expression through "collateral censorship," while compliance costs crush small platforms until only giants remain—immunity actually protects ordinary people's space to speak. For accountability (steelman): immunity lets platforms externalize the social cost of harmful content, amplifying incitement and fraud through recommendation algorithms without bearing the consequences; a measured duty of care forces investment in governance. The key variable is how penalties are designed: the more "one-size-fits-all deadlines and heavy fines," the stronger the over-removal chill; the more it stresses "process diligence," the better it can balance expression and safety—but the harder it is to enforce.

Misconception: "Once a platform edits or removes content, it should answer for everything like a media outlet." Take US Section 230—the opposite is true: its very design is to let platforms moderate confidently (remove spam, block harassment) without being treated as publishers. Others wrongly think immunity "requires platforms to be politically neutral"; the statutory text imposes no such condition. Mistaking the policy preference "should platforms be neutral" for what current law actually says is a common confusion.

First, sort the terms: misinformation is error spread unintentionally; disinformation is deliberately manufactured to mislead. They are often conflated, but their governance logic differs—the former calls for clarification, the latter also for tackling the motive and the coordinated manipulation behind it.

Why is governance so hard? Three structural obstacles: truth is often contested (scientific and political judgments get revised over time; today's "false" may be tomorrow's "right"); who adjudicates (handing "arbiter of truth" power to a government or platform is itself dangerous); and chilling and backfire (a "false" label sometimes makes people believe more strongly—the backfire effect). So mature governance rarely relies on deletion alone.

• Counter-speech (more speech): rooted in the US tradition—the best remedy for bad speech is "more speech," not silencing, diluting lies through open debate and independent fact-checking rather than official verdicts on truth.
• Immunity through education: countries such as Finland embed media literacy in public education, training citizens to spot manipulation themselves; this works slowly, but sidesteps the "who decides what's true" power problem.
• The double edge of "fake news laws": some countries criminalize "spreading false information." Supporters say it curbs malicious rumor; critics warn that "false" is vaguely defined and, historically, easily turned against criticism and dissent—the tool is neutral, the risk lies in whose discretion wields it.

The intervene-vs-abstain tension. For active governance (steelman): coordinated disinformation can manipulate elections, incite violence, and endanger public health; standing aside lets "a lie run round the world while the truth is putting its shoes on," so labeling, throttling, and even legal tools are warranted. For cautious restraint (steelman): once any party is granted the power to "designate disinformation," it will almost inevitably expand and be used to suppress legitimate dissent; and labeling often triggers backfire, so invest in education and transparency instead. The pragmatic consensus lands on the low-coercion end: prioritize fact-checking, media literacy, and algorithmic transparency, reserving "criminalization" for a very narrow, clearly definable set of cases (e.g., deliberate fraud in public-safety domains).

Misconception: "Just delete the false stuff, problem solved." This underrates three things: one, much contested content is opinion or unsettled, not simply-falsifiable "fact"; two, deletion grants someone "arbiter of truth" power that is easily abused to silence dissent; three, the more conspicuous the deletion, the more it confirms a "they're hiding the truth" narrative and amplifies it. "Delete" is the laziest yet most dangerous move—usually not the most effective one.