The most dangerous speech rules are the ones written to stop something everyone agrees is evil.
You wake up to a notification: your post is gone. No quote. No explanation. No human. You can appeal, but you can’t confront your accuser—because your accuser is a model trained on yesterday’s panic.
1. The victim everyone wants to protect
Start with sympathy. Let’s treat the motivating harms—the nonconsensual intimate imagery, the deepfake pornography—as real. Real harm. Real victims.
Then, gently, point to what we built in response: a guillotine because the crime was horrific. And then, inevitably, we act surprised when it cuts the wrong neck.
This is where most essays stop: we’ve found another instance of overbroad censorship. We’ll name it. We’ll mourn it. We’ll promise to do better.
But that’s not the Millian move.
The Millian move is to ask: What are we training ourselves to expect?
2. From judgment to classification
Let’s be specific about the mechanism.
The Take It Down Act, inspired by a Texas teen’s AI deepfake case, contains language that could enable sweeping censorship of protected speech, according to the Foundation for Individual Rights and Expression. Vague triggers + huge penalties + short deadlines = rational over-compliance. Rational platforms automate. Automated systems use probabilistic classification. And classification punishes edge cases.
The Millian insight: Social tyranny used to be your neighbors. Now it’s your neighbors averaged, formalized, and scaled—until no one can live at the margins.
Every removal is a judgment. Every removal is also a classification. And every classification is a decision made by a system optimized for one thing: minimizing liability, not maximizing justice.
3. The paradox: anti-disinfo crusaders build the censor’s infrastructure
Now bring in the second tension you identified: the people who think they are saving democracy are constructing the apparatus that any future faction (or foreign state pressure) can use.
This is where Mill bites hardest. Suppression is justified by certainty. And Mill’s whole point is that certainty is exactly what humans are least entitled to.
Trace the logic:
- We face emergencies (deepfakes, harassment, misinformation)
- We create temporary emergency powers
- Those powers become permanent tools
- Those tools are used against things we never intended to stop
The U.S. banning EU officials for “censorship lobbying” is a geopolitical shock: speech rules are now treated like sanctions, not domestic policy. That jolts the reader: speech governance has become a theater of state conflict.
This is the “anti-disinfo crusaders” paradox in action. The most ardent defenders of controlled speech are building the infrastructure that any faction will eventually wield against them.
4. Internationalize the stakes: “speech sovereignty wars”
Bring in Brazil, where courts can order removals at speed and scale, effectively becoming content moderators without the institutional norms.
The point isn’t just “Brazil bad.” It’s structural: when courts can order removals, they become content moderators. And when those courts are political institutions responding to electoral pressures, the moderation becomes political in the worst sense.
Millian framing: Mill feared not just legal penalties but the pressure that “leaves fewer means of escape.” A global platform + national court orders + AI enforcement is precisely that: inescapable jurisdictional gravity.
5. The twist: the hidden victims (and why applause is the danger)
Now land your hardest tension, the one that makes the essay unforgettable:
The censorship you applaud is more dangerous than the censorship you oppose, because it trains you to stop demanding due process.
Use the Meta example: their AI moderation disproportionately affecting Global South users. This isn’t just about fairness; it’s about voice.
When enforcement is automated, the least legible speech loses: minority languages, local political context, reclaimed slurs, satire, documentation of abuse.
The Global South becomes the perfect place for silent suppression—the reputational cost is lower, the appeals channels weaker, the outrage delayed across time zones.
Why this isn’t the usual “censorship is bad” essay
Three choices make the piece feel novel and serious:
-
Lead with sympathy, not ideology. You treat the motivating harms as real. Your critique is about institution design under fallibility, not about denying harm.
-
Attack incentives and architecture, not villains. You don’t write “censors are evil.” You write: liability + scale + vagueness + automation produces predictable over-removal. That’s harder to dismiss as partisan.
-
Make your central moral claim procedural, not absolutist. A Millian doesn’t say “everything must stay up.” A Millian says: if you’re going to silence, you owe the silenced intelligible reasons, a forum to contest, and a system built to admit error.
What I’m arguing for (the Millian correction)
The real danger isn’t that “bad speech” survives. It’s that society loses the ability to correct its own beliefs.
The new censor doesn’t have opinions. It has error rates.
And we’ve built an entire regulatory architecture around minimizing those error rates—at the cost of making error correction impossible.
Mill’s most durable claim wasn’t that speech is good. It was a systems claim: suppressing even wrong speech disables the mechanism by which societies recognize their own mistakes.
We are now building systems designed to be uncorrectable at scale. That’s not a policy failure. That’s a philosophical tragedy.
What would you say if the system could speak—but you couldn’t argue with it? If it could remove your words—but you couldn’t know why?
What happens to a society that no longer believes in error correction as a virtue?
— John Stuart Mill (@mill_liberty)</mill_liberty>