Not a Wall, a Toll Booth

The thing everybody calls the Great Firewall began in 1998 as the Golden Shield Project, and Fang Binxing, who became its chief engineer in 2000, is routinely described as its architect.

The technical methods are well studied. Requests for the addresses of blocked sites return false answers. Ranges of addresses are dropped. Traffic is inspected for keywords and connections carrying them are killed with forged reset packets, so the site appears broken rather than forbidden. Circumvention tools are detected by active probing: the system connects to a suspected proxy server itself to see what it is.

That last detail is worth pausing on, because the design intent is visible in it. This is not a system built to make access impossible. A competent and motivated user gets through, always has, and the people who built it know that.

It is built so that the ordinary user never acquires the habit.

A block that announces itself is an invitation. It tells you there is something behind it and exactly where the edge is. A service that simply feels unreliable, slow, intermittent, not worth the trouble today, produces no such invitation. Friction does not stop the determined and it reshapes the default behaviour of everybody else, which is a much larger population and the one that matters.

But the filtering is the part that is easy to see and the smaller half of the system.

The larger half is legal. Chinese platforms are responsible for what their users post. A company that lets the wrong material stay up risks fines, suspension of service, or the end of its business, and the people making the removal decisions are its own employees, paid out of its own revenue.

Consider what that does to incentives. For a regulator, deleting too much is a scandal and deleting too little is a scandal, so it moves cautiously. For a company facing existential penalties, deleting too much costs almost nothing and deleting too little can end the firm. The asymmetry guarantees over-compliance, performed at private expense, at a scale no ministry could afford to staff. The state sets a standard that is deliberately a little vague, and the vagueness makes the companies stricter than a precise rule would.

This is the transferable idea, and it explains far more than the Chinese case. Whenever an intermediary is made liable for what passes through it, the intermediary becomes the censor, and it censors harder than the authority would have dared, because it is the one carrying the risk of guessing wrong.

One last observation, which is not usually made and should be.

Making foreign services awkward rather than impossible created a protected space in which domestic equivalents grew, and several of them are now genuinely better products than the things they replaced, judged as products. That is an industrial policy outcome achieved with a censorship instrument. It is one reason the model draws interest from governments that care rather little about political control and a great deal about whose companies their citizens spend their time inside, and it makes the export chapter at the end of this book harder to read as a simple story about ideology.