Europe's Digital Services Act and the First Amendment, Side by Side

Imagine two countries, each of which decides that the internet needs adult supervision. One writes a statute that tells platforms what they must remove and by when. The other writes a single sentence that tells the government what it may not do. Only one of those is a speech code. The other is a speech protection, and the gap between them is the subject of this post.

I want to do this comparison honestly, which means admitting at the start that Europe's lawmakers are not cartoon villains. They are legislators with histories, fears and constituencies, working out how to govern a technology that did not exist when their constitutions were written. Their answers are different from ours, and in my view worse, but they are answers to real questions. A fair comparison takes the question seriously and then lays the answers side by side. It also helps to remember that a regulation titled "Digital Services" has roughly the same romance as a tax form with a mission statement.

The claim, and the method

The claim: the speech regimes of the European Union, Germany and the United Kingdom impose duties and penalties on platforms, and define restricted content, in ways that would not survive the First Amendment. Comparing them with American law shows what that law protects, and Americans should keep it.

The method: four jurisdictions, five questions. For each: what is the main duty, what is the penalty, how is restricted content defined, who decides, and what is the underlying theory? Then a verdict on what each model suits.

One caution before the table. Foreign statutes are long and amended often, and I am describing their main features at the level I am confident of. If you are making a decision that turns on a detail, read the text itself, which is public. The point here is structural.

The four regimes in a table

I am going to walk through the columns, because the table hides what it shows.

Premise one: European law places the first decision on a private company, under threat of a large fine

The structure of all three European models is the same. A statute defines a duty. The duty falls on a private intermediary. A regulator or a court can fine the intermediary for failure. The intermediary has to decide, usually fast, whether a given post is on the wrong side of the line.

Consider the incentive. A platform handling millions of posts cannot adjudicate each one with the care of a trial judge. If the penalty for leaving up something unlawful is large and the penalty for taking down something lawful is small, the rational company takes down anything doubtful. Lawyers call that over-removal, and critics of Germany's 2017 law, which the Digital Services Act has largely superseded since February 2024, raised it from the start: the 24-hour deadline for "manifestly unlawful" content leaves no time for nuance, and the fine for systemic non-compliance is far larger than any cost of deleting a borderline post.

The Digital Services Act is more elaborate than the German law, and its defenders point to features that soften the effect: a notice-and-action process, a right to a statement of reasons for removal, appeals mechanisms, and out-of-court dispute settlement. Those are real. They are also the architecture of a system that expects removals at scale. The DSA's heaviest duties fall on "very large online platforms," those reaching more than 45 million monthly users in the Union, who must assess and mitigate "systemic risks," a category that in the text includes negative effects on civic discourse and electoral processes. A duty to mitigate risks to "civic discourse" is a duty to manage the conversation. Whatever the intent, it hands a regulator a standard so broad that almost any speech can be brought within it. "Mitigate risks to civic discourse" is what you write when you want a power with no edges and a very polite name.

Premise two: the definitions differ, and the definition is where the freedom lives

The EU and German systems anchor "illegal content" in existing law. That is better than free-floating "harmful" content, and I want to give it its due. If what is removable is only what is already unlawful, the system is at least tethered to rules legislators passed.

But the tether is longer than an American would assume. The German Criminal Code, for example, includes offences such as incitement to hatred (Volksverhetzung) and the display of symbols of unconstitutional organisations. Those provisions come out of a particular history. Germany's postwar constitution, the Basic Law, was written by people who had watched a democracy be dismantled by a movement that used free speech as a ladder. Article 5 of the Basic Law protects the expression of opinion, and it also says those rights are limited by the provisions of general laws and by the right to personal honour. A country that lived through the Third Reich can be forgiven for a militant theory of democracy. An American can understand it and still decline to import it.

The American answer to the same history was different. The First Amendment's text is "Congress shall make no law ... abridging the freedom of speech, or of the press." In Brandenburg v. Ohio (1969) the Supreme Court held that advocacy, even ugly advocacy by a Klan leader, is protected unless it is directed to inciting imminent lawless action and is likely to do so. We cover that decision in Brandenburg v. Ohio Protected a Klansman's Speech, and It Protects Yours Too. In Snyder v. Phelps (2011) the Court protected picketing of a soldier's funeral with hateful signs because the speech addressed matters of public concern on public land. See Snyder v. Phelps and the Hardest Free Speech Case Conservatives Ever Defended. Those are cases a European court would likely not have decided as ours did, and the difference is not subtle.

The UK model adds a further feature. The Online Safety Act 2023 imposes duties not only about illegal content but about content that is harmful to children, backed by age-assurance obligations. The bill as first introduced also included a category of "legal but harmful" content for adults, which attracted heavy criticism for sweeping lawful speech into a regulatory net, and that adult category was dropped before the Act passed. That retreat is itself instructive: even in a Parliament willing to regulate, the idea that a regulator should police lawful adult speech met resistance. The final law still leaves a large role for the regulator, Ofcom, to write codes and enforce duties, which means that much of the real content of the regime is written after the statute passes, by an agency.

Premise three: who decides is the whole question

In the American system, the first and usually the last decision is made by a court, with the government carrying the burden, and a prior restraint, an order barring speech before it is said, is the least favoured remedy known to First Amendment law. Near v. Minnesota (1931) and New York Times Co. v. United States (1971) stand for that.

In the European models, the first decision is made by a company, applying a legal standard it did not write, under a deadline, with a regulator watching. The courts remain available afterward. But the practical weight of the system rests on the platform's call. This is not an accident. It is the reason the model scales. Courts are slow, and moderation at the scale of the internet must be fast.

That is the honest case for the European approach, and the point at which it is strongest: speed and scale. A system built on a rule that only courts may restrict speech will tolerate a great deal of harmful content for a long time. The Europeans decided that trade-off was unacceptable. Americans, for the better part of two centuries, have decided the opposite.

Which is right? Here the question turns on a judgment about trust. The European model trusts regulators and platform compliance departments to draw the line sensibly and gives citizens recourse if they do not. The American model trusts citizens to cope with a great deal of unpleasantness and falsehood and distrusts anyone, in a government or a company acting for one, who is given the power to decide what they may hear. I think the American wager is better, for a simple reason: the power to censor is a power that will be exercised by people who are interested parties.

Premise four: the American system is not a free-for-all, and Section 230 is not what people say it is

A fair comparison needs a fair account of American law. Platforms in the United States are not helpless against bad content, and the law does not force them to carry everything.

Section 230(c)(1) of the Communications Decency Act of 1996 says that no provider of an interactive computer service shall be treated as the publisher or speaker of information provided by another content provider. Section 230(c)(2) protects good-faith moderation. The combination means a platform may remove what it dislikes and is not generally liable for what its users say. We take the history apart in The 1995 Prodigy Lawsuit That Accidentally Created Section 230 and the myths around it in Publisher or Platform, Eight Myths About Big Tech and the Law.

In Moody v. NetChoice (2024), the Supreme Court addressed state laws in Texas and Florida that sought to limit platforms' ability to moderate. The Court sent the cases back for a more careful analysis, but its opinion made clear that when a platform curates a feed, it engages in expressive activity that the First Amendment protects. See Moody v. NetChoice and the Unfinished Question of Who Owns the Feed. That means American law protects the platform's own speech rights, not only the user's. This is the most important asymmetry in the table: in Europe the company is a regulated duty-bearer; in America it is also a speaker.

American law does restrict some online content: child sexual abuse material, true threats, fraud, and incitement are unprotected. The difference is not that Americans permit everything. It is that American law lists what is unprotected, in categories narrow enough that a court can apply them, and puts the burden on the government.

The best case for the European approach, and the answer

The strongest argument for the European model deserves a full statement. Online platforms are not newspapers. They are infrastructure, used by billions, with algorithms that amplify some speech and bury other speech. A rule that only forbids government from acting leaves private concentrated power unchecked, and a democracy has a legitimate interest in the information ecosystem on which it depends. Europe's regulators are accountable to parliaments and courts, and the rules are written down, which is more than can be said for the unwritten decisions of a few executives in California.

The answer has three parts.

First, concentrated private power over speech is a real problem, and the American response to it is competition, transparency and state-law experimentation within First Amendment limits, not a federal duty to remove content the government dislikes. A company that moderates badly can be left, criticised, and out-competed. A regulator that moderates badly cannot be left.

Second, "accountable to parliament" is not the same as "limited by a constitution." The point of a written bill of rights is that some questions are not up for majority vote. In the American design, no majority, however sincere, can give a regulator the power to decide which lawful opinions are too risky for the public.

Third, the European approach exports. A platform that must meet a European standard for takedowns will often apply a version of it everywhere, because running separate systems is expensive. The result is that European rules shape what Americans can see and say online. That is reason for Americans to understand the comparison and to be wary of the argument that these laws only affect Europeans.

What each model suits

A fair verdict gives each system its due.

The European model suits a political culture that has chosen a "militant democracy," in which the state takes a positive role in protecting the public sphere and where regulators are trusted to apply broad standards. If a society's historical memory is dominated by the failure of an open democracy to protect itself, it is rational to prefer a system that errs toward removal.

The American model suits a political culture that remembers different failures: the Sedition Act of 1798 (see The Sedition Act of 1798 Jailed a Congressman Who Won Reelection From His Cell), the licensing of printers, the prosecution of dissenters in wartime, and the long record of officials who found criticism of themselves to be dangerous. If your historical memory is dominated by the abuse of the censor's pen, it is rational to put the pen out of reach.

I will say what I think. The American model is better, not because Americans are wiser than Europeans, but because the power to decide what is dangerous speech is a power that is always, eventually, used against someone who was right. Our system accepts a higher cost in tolerated ugliness in exchange for a lower risk of a state-run orthodoxy. I would make that trade again.

The case in brief

1. The EU Digital Services Act, Germany's 2017 Network Enforcement Act and the UK Online Safety Act all place removal duties on platforms, enforced by large fines, with a regulator supervising.
2. Their definitions rest on illegal content under national law and, in the UK, content harmful to children, and the duty to manage "systemic risks" gives regulators broad scope.
3. In the American system, a court decides first, the government bears the burden, prior restraint is disfavoured, and the unprotected categories are narrow and old.
4. American law also protects the platform's own editorial choices, as Moody v. NetChoice (2024) recognised, and Section 230 shields moderation without compelling it.
5. The European model fits a militant-democracy tradition, the American model fits a tradition of distrust of the censor, and Americans should keep theirs.

What you can do: Read the First Amendment in the Constitution, then use the debate desk to practise stating the European case at its strongest before you argue against it.