How Internet Censorship Works
When people hear the word censorship, they imagine China.
They imagine government officials staring at computer screens, blocking websites, deleting posts, banning words, and arresting dissidents. The state is visible. The censor has a face.
That is not how censorship usually works in the United States and much of Western Europe.
The West has evolved something arguably more sophisticated. The censor is not a bureaucrat. The censor is liability.
Nobody needs to order Google what to rank. Nobody needs to instruct ChatGPT what subjects to discuss. Nobody has to issue a list of forbidden websites. Instead, an elaborate legal environment encourages platforms to censor themselves before anyone ever files a lawsuit.
The mechanism is simple.
Imagine operating a search engine used billions of times every day. Or an AI system answering millions of questions every hour.
Every answer carries legal risk.
If one answer contributes—even indirectly—to someone's death, financial loss, medical injury, or criminal conduct, lawyers may ask whether the platform bears responsibility. Even if the company ultimately wins in court, years of litigation can cost millions. If it loses, damages may reach hundreds of millions or more.
From the perspective of corporate management, the rational response is obvious:
Avoid risk.
This is not ideological.
It is actuarial.
The Internet that once rewarded openness increasingly rewards caution. Search engines and AI systems therefore gravitate toward information least likely to create legal exposure. The safest answer is usually not the most controversial, the most interesting, or even the most complete. It is the answer that can most easily be defended before a judge.
That incentive quietly reshapes the information ecosystem.
Consider suicide. Twenty years ago, searching the Internet produced forums, philosophical discussions, academic debates, advocacy groups, personal stories, prevention organizations, and yes—even websites openly discussing methods.
Today, much of that material has become difficult or impossible to discover through mainstream search engines. Instead:
- Official prevention organizations dominate.
- Government advice dominates.
- Medical institutions dominate.
The obvious explanation is that technology companies suddenly developed a deep moral concern. A more plausible explanation is considerably less romantic.
Suppose a grieving family argues that an online recommendation, algorithm, chatbot, or search result encouraged a suicide. Whether that claim ultimately succeeds is almost beside the point. The litigation itself becomes punishment.
Recent lawsuits illustrate the legal climate in which platforms operate. Families have sued technology companies alleging that recommendation algorithms contributed to harms suffered by children, including exposure to dangerous content and, in some cases, suicide. While courts have often wrestled with protections such as Section 230 of the Communications Decency Act and questions of causation, the cases demonstrate that platforms face persistent legal pressure over what their systems recommend and amplify.
Medical information provides another illustration. Search for treatments for almost any disease. The first results overwhelmingly come from government agencies, major hospitals, universities, or established medical organizations. Alternative views, minority hypotheses, or unconventional practitioners may still exist somewhere online, but they occupy the margins.
This does not necessarily prove that official advice is always correct. Science progresses precisely because established consensus is sometimes overturned. But if you operate a multi-billion-dollar platform, the safest legal strategy is obvious. If your search engine recommends advice from the Mayo Clinic or the National Institutes of Health, your lawyers have an easier day in court than if you promote an obscure independent researcher whose claims later prove harmful.
The platform's objective is not necessarily truth.
Its objective is legal defensibility.
The same logic extends into artificial intelligence. Every answer an AI generates potentially creates liability. The safest model therefore learns not merely what is likely to be true, but what is safest to say.
Entire categories of responses become wrapped in caveats:
- Medical questions receive reminders to consult physicians.
- Legal questions recommend lawyers.
- Financial questions urge professional advisers.
- Mental health discussions encourage crisis services.
These disclaimers are often sensible. But they also reveal the underlying incentive structure. They exist because someone, somewhere, might someday sue.
This is not a conspiracy.
No secret committee coordinates these outcomes.
No ministry distributes censorship directives.
Instead, thousands of lawyers, insurers, compliance officers, risk managers, and corporate counsel independently arrive at remarkably similar conclusions:
- Reduce exposure.
- Reduce uncertainty.
- Reduce liability.
The cumulative effect resembles censorship even though no single actor intended to censor.
- Economists call this an incentive problem.
- Lawyers might call it prudent risk management.
- Citizens may simply experience it as disappearing information.
Several high-profile court battles illustrate how legal pressure shapes online behavior even when platforms prevail. In Gonzalez v. Google (2023), the U.S. Supreme Court considered whether YouTube's recommendation algorithms could expose Google to liability under anti-terrorism law. The Court ultimately resolved the case on related grounds without redefining Section 230, but the litigation highlighted how recommendations themselves have become legal targets. Likewise, numerous suits against social media companies over alleged harms to minors have focused not only on user-generated content but on the platforms' design and recommendation systems.
Whether these claims succeed is almost secondary. The mere possibility that they might succeed changes corporate behavior.
Risk departments do not wait for final judgments. They anticipate them.
Every lawsuit sends a signal.
Every settlement sends another.
Eventually, engineers modify algorithms not because governments demand it, but because lawyers recommend it.
This is how a free society can gradually narrow the range of easily accessible information without ever establishing a Ministry of Truth:
- The state writes laws.
- Private litigation enforces incentives.
- Corporations respond rationally.
- Algorithms quietly adjust.
- Users notice only that certain information has become harder to find.
None of this means every lawsuit lacks merit, nor that platforms should never remove dangerous content. Companies have legitimate responsibilities, and there are compelling reasons to reduce the spread of fraud, exploitation, and material that poses an immediate risk of serious harm.
But it does mean we should be honest about the trade-offs.
When legal risk becomes sufficiently large, information is filtered not only according to its accuracy but according to its litigation potential.
The result is not censorship in the classic authoritarian sense. It is something subtler.
Nobody tells platforms what they must say.
The legal environment tells them what they cannot afford to say.
That distinction may satisfy constitutional lawyers. For ordinary citizens trying to understand the world, it often feels like a distinction without much difference.