Key findings, August 2026
- Not one country did both. Of the 23 countries that spent at least 5% of the month with a measured outage, none also blocked a platform at national scale. Of the 8 countries that block at national scale, none lost meaningful connectivity.
- The most disrupted country blocked nothing. Nigeria spent 76% of the month with a measured outage and has 0 platforms confirmed blocked out of 25.
- The heaviest blocker stayed online. Iran blocks 21 of 25 platforms and spent 2.8% of the month disconnected. Filtering requires a working network.
- An outage hides the evidence, but only for some. Of the 16 most disrupted countries, 7 are absent from the blocking dataset entirely and 3 more have coverage too thin to judge. The rest are well measured — which is what keeps the finding from resting on missing data.
Free to quote with attribution to vpnrank.io and to both IODA and OONI as data sources. Please state the measurement window.
Two different events share one word, and the word hides the difference
When a country is described as having an internet shutdown, the reporting rarely distinguishes between a network that stopped carrying traffic and a network that is carrying traffic but refusing particular destinations. The first is usually infrastructure — a cut cable, a power failure, a conflict, a routing withdrawal. The second is policy, and it requires the network to keep working in order to work at all.
The distinction is not academic. It changes who is responsible, what fixes it, and what a person inside the country can do about it. Nothing installed on a phone helps when the routes are gone. This page separates the two by measuring them separately and putting them on the same row.
The two lists barely intersect
Ranking by the share of the month a country spent with a measured outage, and setting each country’s blocking figure beside it, produces two populations that hardly overlap. The countries losing connectivity are largely in West and Central Africa, the Caribbean and the Pacific; the countries blocking platforms are elsewhere entirely.
| Country | Share of month offline | Longest single outage | Platforms blocked | What it doesn't automatically prove |
|---|---|---|---|---|
| Nigeria | 76% | 78h | 0 of 25 | Only 11 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Angola | 56% | 418h | 0 of 25 | Coverage here is good, so the zero is a measured result rather than a gap. It still says nothing about blocking that OONI cannot confirm, such as DNS tampering. |
| Côte d’Ivoire | 45% | 336h | 0 of 25 | Coverage here is good, so the zero is a measured result rather than a gap. It still says nothing about blocking that OONI cannot confirm, such as DNS tampering. |
| British Virgin Islands | 43% | 321h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
| Mali | 42% | 309h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
| Syria | 38% | 200h | 0 of 25 | Only 13 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Guatemala | 35% | 256h | 0 of 25 | Coverage here is good, so the zero is a measured result rather than a gap. It still says nothing about blocking that OONI cannot confirm, such as DNS tampering. |
| Cape Verde | 32% | 15h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
| Tunisia | 26% | 90h | 0 of 25 | Only 11 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Northern Mariana Islands | 25% | 16h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
| Solomon Islands | 25% | 180h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
| American Samoa | 17% | 126h | not in the blocking dataset | Absence from the blocking dataset is absence of measurement, not absence of blocking. Nothing can be concluded about this country either way. |
Reading the same join from the other end makes the point harder. Every country that blocks platforms at national scale kept its network up.
| Country | Platforms blocked | Share of month offline | What it doesn't automatically prove |
|---|---|---|---|
| Iran | 21 of 25 | 2.81% | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
| Russia | 12 of 25 | 0.08% | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
| Germany(flagged) | 8 of 25 | 0.07% | This country's confirmed blocks sit only on VPN and Tor domains with social and news clean — the fingerprint of a corporate or campus filter inside the probe pool, not a state blocklist. It is flagged and excluded from the censorship ranking. |
| Mexico(flagged) | 8 of 25 | 0.17% | This country's confirmed blocks sit only on VPN and Tor domains with social and news clean — the fingerprint of a corporate or campus filter inside the probe pool, not a state blocklist. It is flagged and excluded from the censorship ranking. |
| Myanmar (Burma) | 2 of 25 | 0.74% | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
| Türkiye | 2 of 25 | no outage observed | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
| India | 1 of 25 | no outage observed | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
| Thailand | 1 of 25 | 0.09% | That a blocking country has no outage does not mean its network is healthy everywhere, only that no outage large enough to register was observed. |
Practical ruleBefore treating a reported shutdown as censorship, ask which instrument saw it. A routing withdrawal and a blocked domain are measured by different tools and mean different things — and a country doing the second needs its network to stay up.
The countries that go dark are the ones we can say least about
There is a limit built into this join that has to be stated before the finding can be trusted. Blocking measurements come from volunteer probes running inside the country. When the network goes down, those probes stop reporting — so the very event under study suppresses the evidence about everything else happening on that network.
The effect is real, and it is worth being precise about its size rather than dramatic. Of the 16 countries that lost at least 10% of the month, 7 do not appear in the blocking dataset at all. Among those that do, 3 have coverage too thin to draw anything from — Nigeria on 11 of 25, Syria on 13 of 25, Tunisia on 11 of 25 — while the median country was judged on 24 of 25. So the shadow falls on a minority, not on the whole set. Reading “nothing blocked” off one of those thin countries would still be exactly the error this data invites.
The finding survives it because of the control cases. Angola (56% offline, 24 of 25 measurable); Guatemala (35% offline, 25 of 25 measurable); Cameroon (12% offline, 25 of 25 measurable); Somalia (11% offline, 25 of 25 measurable); Ethiopia (11% offline, 24 of 25 measurable) each lost a substantial share of the month and kept nearly full measurement coverage, and each blocked nothing. The pattern therefore also appears where the probes stayed up, which is what distinguishes it from an artefact of missing data.
Method, and the three ways this data misleads
Outage data comes from IODA’s event feed for the window 2026-08-01 to 2026-09-01, paged in full (603 events). Blocking data comes from our own monthly censorship index, built from OONI measurements over the same window. Every figure here is recomputed when either source is re-pulled.
| The trap | What it does to the numbers | What we do instead |
|---|---|---|
| IODA’s score is not comparable across countries | In this window the score runs from 2,274,045 down to double digits. Most of that spread is the size of the country's address space, not the severity of the disruption. Ranking on it ranks by country size. | Every figure is a share of the measurement window, so a small country and a large one sit on the same scale. |
| Event counts point the wrong way | A country can record dozens of brief events and less total disruption than a country with two long ones. Counting events measures fragmentation, not severity. | Event counts are shown for context and never ranked on. |
| Durations double-count if summed | Events overlap the window edges — 11 did here — and the same real outage is reported separately by each datasource, since routing withdrawal and active probing observe it differently. | Events are clipped to the window, then intervals are unioned rather than summed, per datasource and then across them. |
Limitations. A month is a coarse window: a country that lost connectivity for six hours during an election looks similar here to one with chronic infrastructure problems, and the longest-single-outage column is the only correction offered. Both datasets under-observe countries with few probes and few monitored networks, which biases against exactly the places most likely to experience either event. Coincidence within a month is not causation and nothing here should be read as one thing causing the other. And a country with no measured outage is a country where no outage was observed — not one where none occurred.
Sources, licence and citation
Outage measurements are collected and published by IODA, the Internet Outage Detection and Analysis project at the Georgia Institute of Technology, which combines BGP routing data with active probing and background traffic analysis. Their data carries the notice “Copyright (c) 2021-2025 Georgia Tech Research Corporation. All Rights Reserved.” — what we publish here are derived per-country shares with attribution, not their event feed.
Blocking measurements come from OONI, the Open Observatory of Network Interference, published openly under a Creative Commons licence. Neither project has reviewed or endorsed this analysis; the join, the thresholds and every judgement about what the numbers do and do not support are ours.
Reuse and citation
Citation: vpnrank.io. A blackout is not a block. Analysis of IODA outage and OONI web connectivity data, August 2026 edition. Accessed [date]. https://vpnrank.io/research/blackout-is-not-a-block
This URL always holds the current edition. Credit IODA and OONI alongside us, and state the window — a shutdown figure without its window is not a fact.
Request the underlying cut or a specific country →How we are funded
vpnrank.io earns affiliate commission when readers subscribe to VPN services we cover. No provider is named or ranked anywhere in this analysis, and neither dataset here has anything to do with the products we earn from. The method is published so the arithmetic can be checked against IODA’s and OONI’s own APIs.
Related: the monthly censorship index holds the platform-by-country blocking data this page joins against, and the research hub lists everything else we publish with a documented method.