Two ways to censor a network: VPN-blockers vs news-blockers

Most of what we call “censorship” is measured as a single number: how much a country, or a network, blocks. But volume hides intent. Two networks can block the same amount and be doing completely different things — one trying to stop people from reading, the other trying to stop them from getting around it. Voidly’s per-ISP category data (/v1/measurement/isp-categories) lets us see the shape of a network’s blocking, not just its size — and the shape is a fingerprint of what the censor is actually afraid of.

Most networks block everything they can

Across 86 networks with enough categorised measurements to judge (at least three content categories and five confirmed blocks), the common pattern is blunt: block across the board. Venezuela’s state operator AS8048 is the archetype — it blocks 100% of measured domains in every one of nine categories: news (158/158), gambling (182/182), communication tools (96/96), circumvention tools (90/90), and on down the list. There is no target here, just a wall.

But a handful of networks choose — and reveal two opposite strategies

Only 8 of the 86 show a clearly selective profile (a gap of 40+ points between their most- and least-blocked category). In that minority, two mirror-image strategies stand out:

These are opposites. KZ AS207446 blocks 94% of the anti-censorship tools it sees but leaves news and gambling untouched — a network whose censorship is about stopping people from routing around it, not about any particular content. RU AS47541 does the reverse: it blocks 100% of measured news and 100% of communication tools, but only 9% of circumvention tools — a network about controlling the message, leaving the escape hatch comparatively open.

Others sit in between but still tip their hand. Tanzania’s AS37349 blocks 91% of gambling but only 10% of communication tools — a morality filter, not a political one. Myanmar’s AS9988 blocks gambling (71%) and comms (61%) heavily while leaving AI/ML services (0%) and streaming (8%) alone.

Why the shape matters

The blocking shape tells you the censor’s priority — and, practically, what a user can do about it. In a news-blocker network like RU AS47541, a VPN probably still works (only 9% of circumvention tools are blocked), so the censorship is evadable. In a VPN-blocker network like KZ AS207446, the same VPN probably won’t connect — the network has decided that circumvention itself is the threat. A volume-only ranking would miss this entirely; both could read as “moderate blocking.”

Honest caveats

• Block rate is over measured domains per category, not the whole web. The denominator is the categorised measurements Voidly has for that network; a category that does not appear is unmeasured, not “allowed.”
• Small denominators on some cells. KZ AS207446’s news and gambling rates rest on just six measurements each; its circumvention signal (n=49) is the robust one. The RU contrast is well-sampled on both sides (news n=64, circumvention n=54). Always read the per-category measured count.
• Coverage skews toward blocked endpoints. ASN-tagged evidence is dominated by CensoredPlanet DNS measurements, which over-represent networks where something was blocked — which is exactly why most of the 86 read as “blanket” and selective detection here is conservative, a floor on how much targeting exists.
• This is an observation of measured blocking shape, not an attribution of stated policy to a named operator; ASNs can change hands and routing.

Reproduce any of this live: /v1/measurement/isp-categories?country=KZ&asn=207446 (selective), ?country=RU&asn=47541 (the opposite), or ?country=VE&asn=8048 (blanket). Pass just ?country=XX for a country’s networks ranked.