---
title: "Internet Shutdowns vs Website Blocking — What the Measurements Show (September 2026)"
description: "Measured connectivity outages and measured platform blocking, joined country by country for September 2026. The countries that lose the internet and the countries that block platforms turn out to be different countries. Updated monthly."
url: https://vpnrank.io/research/blackout-is-not-a-block
language: en
updated: 2026-10-11
image: https://vpnrank.io/enrich/blackout-is-not-a-block/outage-vs-blocking.png
---

vpnrank.io research

# A blackout is not a block

In September 2026, none of the 23 countries that spent at least 5% of the month with a measured connectivity outage also blocked a platform at national scale. “Internet shutdown” is used for two different events: a country losing connectivity, and a country blocking specific platforms while staying online. They have different causes and different remedies. We measured both for the same month and joined them country by country to see how often they are the same thing.

Updated monthly Current edition: **September 2026** · data pulled 2026-10-11

Connectivity outages

[IODA](https://ioda.inetintel.cc.gatech.edu/) — Georgia Tech. 786 events, 123 countries.

Platform blocking

[OONI](https://ooni.org/) — 25 platforms, 179 countries.

Window

September 2026 (`2026-09-01` to `2026-10-01`, end exclusive)

Analysis

[Diego Pereyra](https://vpnrank.io/authors/diego-pereyra), Security & Protocols Specialist

## Is an internet blackout the same as website blocking?

No. A blackout is a network that stops carrying traffic; a block is a working network refusing particular platforms, and in the measurements they happen to different countries. In September 2026, none of the 23 countries with a measured outage covering at least 5% of the month blocked a platform at national scale, and the 7 countries that did block (8 counting 1 flagged as possible filtered-network artefact) were offline for at most 3.6% of it (India).

## Does a VPN work during an internet shutdown?

Not during a blackout: a VPN needs a working connection to tunnel through, and when the routes are gone there is nothing for it to use. It can only help with the other event, platform blocking, which needs the network to stay up: in September 2026, Iran blocked 21 of the 25 platforms we measure while offline for 1.7% of the month.

## Which country lost the most internet connectivity in September 2026?

Cape Verde spent 93% of September 2026 with a measured connectivity outage, the largest share of the 123 countries with any outage in IODA's data, and it does not appear in the blocking dataset, so nothing can be said about its platform blocking. The next most disrupted were Nigeria (83%) and Syria (66%).

Outage share is the fraction of the month with an outage detected by IODA, unioned across its datasources so nothing is counted twice; it does not say what caused the outage. Blocking means OONI-confirmed at national scale, and countries flagged as possibly measured on a locally filtered network are not counted. Data for September 2026, pulled 2026-10-11.

![Illustration, not data: an isometric city split down the middle — one half unlit with severed network cables hanging loose, the other half fully lit but with several doorways sealed shut by amber barriers](https://vpnrank.io/enrich/blackout-is-not-a-block/blackout-hero.png)

Illustration. The two events this page separates: a network that stopped carrying traffic, and a network carrying traffic that refuses particular destinations.

Key findings, September 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.** Cape Verde spent 93% of the month with a measured outage and has platforms confirmed blocked out of 25.
-   **The heaviest blocker stayed online.** Iran blocks 21 of 25 platforms and spent 1.7% of the month disconnected. Filtering requires a working network.
-   **An outage hides the evidence, but only for some.** Of the 18 most disrupted countries, 8 are absent from the blocking dataset entirely and 8 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](https://ioda.inetintel.cc.gatech.edu/) and [OONI](https://ooni.org/) 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.

![Two ranked lists side by side for August 2026: countries that lost the most connectivity (Nigeria 76%, Angola 56%, Côte d'Ivoire 45%, Syria 38%, Guatemala 35%) each blocking 0 of 25 platforms, and countries blocking the most platforms (Iran 21, Russia 12) each losing under 3% of the month](https://vpnrank.io/enrich/blackout-is-not-a-block/outage-vs-blocking.png)

The same month, measured two ways. Verified September 2026.

## 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 |
| --- | --- | --- | --- | --- |
| Cape Verde | 93% | 225h | 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. |
| Nigeria | 83% | 275h | 0 of 25 | Only 11 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Syria | 66% | 475h | 0 of 25 | Only 13 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Tunisia | 59% | 205h | 0 of 25 | Only 13 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Northern Mariana Islands | 52% | 88h | 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. |
| Paraguay | 51% | 366h | 0 of 25 | Only 11 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Laos | 47% | 336h | 0 of 25 | Only 15 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Tonga | 47% | 336h | 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. |
| St. Pierre & Miquelon | 45% | 314h | 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. |
| Bermuda | 44% | 317h | 0 of 25 | Only 15 of 25 platforms had enough measurements here. A low blocking figure from thin coverage is weak evidence. |
| Antarctica | 39% | 279h | 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. |
| São Tomé & Príncipe | 33% | 236h | 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. |

Countries with a measured outage covering at least 10% of September 2026, with their platform-blocking figure alongside. Outage data: IODA. Blocking data: OONI. Pulled 2026-10-11.

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 | 1.71% | 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. |
| Finland | 12 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. |
| Russia | 12 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. |
| Mexico(flagged) | 3 of 25 | 1.62% | 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 | 0.97% | 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 | 3.61% | 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 | 2.82% | 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 | 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. |

Every country with platform blocking above the national-scale threshold in September 2026, with its measured outage share. Pulled 2026-10-11.

> 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

![Illustration, not data: an isometric dark terrain covered in measurement probes, most of them switched off and grey, a handful still sending a thin beam of light upward](https://vpnrank.io/enrich/blackout-is-not-a-block/probes-going-quiet.png)

Illustration. The measured version of this is the chart below.

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.

![Bar chart of the most disrupted countries in August 2026 with how many of 25 platforms were measurable in each: Nigeria 76% offline with only 11 of 25 measurable, Syria 38% with 13 of 25, while Guatemala 35%, Cameroon 12% and Somalia 11% had all 25 measurable](https://vpnrank.io/enrich/blackout-is-not-a-block/measurement-shadow.png)

The measurement shadow, and the three countries that escape it. Verified September 2026.

The effect is real, and it is worth being precise about its size rather than dramatic. Of the 18 countries that lost at least 10% of the month, 8 do not appear in the blocking dataset at all. Among those that do, 8 have coverage too thin to draw anything from — Nigeria on 11 of 25, Syria on 13 of 25, Tunisia on 13 of 25, Paraguay on 11 of 25, Laos on 15 of 25, Bermuda on 15 of 25, Lithuania on 11 of 25, South Sudan on 15 of 25 — while the median country was judged on 15 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. Somalia (27% offline, 25 of 25 measurable); Guatemala (14% 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 2026-09-01 to 2026-10-01, paged in full (786 events). Blocking data comes from our own monthly [censorship index](https://vpnrank.io/research/internet-censorship-map), 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 900,468 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 — 12 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. |

The three normalisation errors this dataset invites, and the handling written into the collector. Thresholds and rules are recorded in the published data file so any figure here can be audited against it.

**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](https://ioda.inetintel.cc.gatech.edu/) 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](https://ooni.org/), 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, September 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 →](https://vpnrank.io/press)

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](https://vpnrank.io/research/internet-censorship-map) holds the platform-by-country blocking data this page joins against, and the [research hub](https://vpnrank.io/research) lists everything else we publish with a documented method.

Structured data (JSON-LD):

```json
{"@context":"https://schema.org","@type":"Organization","@id":"https://vpnrank.io/#organization","name":"VPNRank","alternateName":"vpnrank.io","url":"https://vpnrank.io","email":"hello@vpnrank.io","logo":{"@type":"ImageObject","url":"https://vpnrank.io/logos/favicon.png","width":500,"height":500},"description":"Independent VPN comparison. We track 9 major VPN services every day and rank them for privacy, security, speed and value.","foundingDate":"2026-03","sameAs":["https://www.youtube.com/@vpnrank","https://www.instagram.com/vpnrank.io","https://www.facebook.com/people/VPNRankio/61592338017872/"],"knowsAbout":["Virtual Private Networks","Internet Privacy","Cybersecurity","Network Encryption","VPN Protocols"]}
```

```json
{"@context":"https://schema.org","@type":"Organization","@id":"https://vpnrank.io/#editorial-team","name":"VPNRank Editorial Team","url":"https://vpnrank.io/about","parentOrganization":{"@id":"https://vpnrank.io/#organization"},"description":"A team of privacy and cybersecurity analysts who independently test VPN services for speed, security, leak protection, and value.","knowsAbout":["VPN testing methodology","WireGuard and OpenVPN protocols","DNS and IPv6 leak testing","No-logs policy audits","Encryption standards"]}
```

```json
{"@context":"https://schema.org","@type":"WebSite","@id":"https://vpnrank.io/#website","name":"VPNRank","alternateName":"vpnrank.io","url":"https://vpnrank.io","publisher":{"@id":"https://vpnrank.io/#organization"},"inLanguage":"en-US"}
```

```json
{"@context":"https://schema.org","@type":"Article","headline":"A blackout is not a block: measured outages against measured blocking (September 2026)","description":"Measured connectivity outages and measured platform blocking, joined country by country for September 2026. The countries that lose the internet and the countries that block platforms turn out to be different countries. Updated monthly.","datePublished":"2026-09-02","dateModified":"2026-10-11","url":"https://vpnrank.io/research/blackout-is-not-a-block","mainEntityOfPage":{"@type":"WebPage","@id":"https://vpnrank.io/research/blackout-is-not-a-block"},"image":"https://vpnrank.io/enrich/blackout-is-not-a-block/outage-vs-blocking.png","author":{"@type":"Person","name":"Diego Pereyra","url":"https://vpnrank.io/authors/diego-pereyra","jobTitle":"Security & Protocols Specialist"},"publisher":{"@type":"Organization","name":"vpnrank.io","url":"https://vpnrank.io"},"isBasedOn":[{"@type":"Dataset","name":"IODA outage events","url":"https://ioda.inetintel.cc.gatech.edu/","creator":{"@type":"Organization","name":"IODA, Georgia Institute of Technology"}},{"@type":"Dataset","name":"OONI web connectivity measurements","url":"https://ooni.org/data/","creator":{"@type":"Organization","name":"Open Observatory of Network Interference (OONI)"}}]}
```