voidly
Auto-generated · refreshes every 6h

Findings — surprising signals in our censorship dataset

14 headline-grade research findings, mined automatically from 2,636 citable incidents and 85,389 underlying measurements across 130 countries. Every finding cites the API call it came from. Free under CC BY 4.0.

14
live findings
2,636
incidents
85K
evidence rows
130
countries

generated Thu, 21 May 2026 04:12:33 GMT · earliest record: 2016-07-15

New report · April 2026

AI Services Censorship Report — April 2026

ChatGPT and Gemini are hard-blocked in China; Claude is reachable. Russia interferes with all five major Western AI services. 3,604 OONI events across 74 countries.

Read the report →

Top findings

Each card has a permalink and a copyable BibTeX/RIS citation. Numbers are computed at request time from public APIs.

Subscribe via RSS
F01Methodology#

2,636 citable censorship incidents documented across 130+ countries

Voidly maintains the largest open, machine-readable corpus of internet censorship incidents — 85.4K pieces of evidence linked to specific outages, blocking events and disruptions, each with a permanent ID and citation.

2,636
incidents
85.4K
evidence rows
+82
new this week
+2055/day
evidence growth
How we computed this

Counted from the live incidents table on api.voidly.ai. "Evidence rows" are the underlying OONI / IODA / CensoredPlanet measurements linked to each incident.

$ curl "https://api.voidly.ai/data/incidents/stats"
GET /data/incidents/statsView open data hub →
F02Events#

Newest detected event: Venezuela — Internet connectivity disruption in Venezuela

Voidly's pipeline ingests new evidence every two hours. The most recent incident in the database started 7h ago (2026-05-20T20:30:00.000Z). Severity: critical; sources: ioda.

7h ago
detected
VE-2026-0215
incident id
critical
severity
ioda
sources
How we computed this

Sorted live incidents API by start_time descending, took row 0. Time-ago is computed at request time.

$ curl "https://api.voidly.ai/data/incidents?limit=1"
GET /data/incidents?limit=1Open incident →
F03Censorship#

Russia leads global censorship rankings (score 33 / 100); 0 countries classified high or severe

Voidly's continuous live index ranks 130 countries on a 0-100 censorship score, derived from network-level test outcomes (DNS, TCP, TLS, HTTP). Countries scoring 70+ are classified "severe", 50-69 "high".

0
high or severe
33
#1 Russia
31
#2 China
24
#3 Iran
How we computed this

Score is a weighted blend of block rate, breadth (how many distinct services), method severity, and network-test agreement across OONI / CensoredPlanet / probes.

$ curl "https://api.voidly.ai/data/censorship-index.json"
GET /data/censorship-index.jsonView full index →
F04Predictive#

27 countries currently exceed Voidly's 7-day shutdown-risk threshold; Iran sits highest

Voidly Sentinel's XGBoost forecast (AUC 0.98, conformal coverage 0.905) projects per-country probability of a major connectivity event in the next 7 days. Threshold-crossing countries trigger HMAC-signed webhook alerts to subscribers within minutes.

87.0%
Iran 7-day risk
87.0%
Iran
82.7%
Kazakhstan
82.6%
Nigeria
82.1%
Venezuela
How we computed this

XGBoost classifier trained on 1.6M historical OONI records + IODA outage labels + event calendar. Output is calibrated via isotonic regression; threshold is set to the 90th percentile of historical false-positive rates.

$ curl "https://api.voidly.ai/v1/sentinel/global_heatmap?min_risk=0"
GET /v1/sentinel/global_heatmapSee risk briefings →
F05Platforms#

TikTok is the most-targeted platform globally — blocking evidence in 40 countries

Voidly aggregates per-platform measurements across 12 services (messengers, social, multimedia, search). Twitter / X consistently leads as the most-blocked service, followed by encrypted messengers Signal and Telegram.

40
countries with evidence
40
TikTok
38
Twitter / X
36
WhatsApp
30
Signal
How we computed this

Counted as DISTINCT country_code in evidence rows where signal_type indicates blocking and the test target maps to the platform. A platform is "blocked" in a country if at least one measurement shows interference.

$ curl "https://api.voidly.ai/v1/platforms/scores"
GET /v1/platforms/scoresSee full platform table →
F06Privacy#

Encrypted messaging is under pressure: 36+ countries show blocking of Signal, WhatsApp or Telegram

Across the three major end-to-end-encrypted consumer messengers, Voidly has found blocking evidence in dozens of countries. Telegram leads as the most-blocked, followed by Signal and WhatsApp.

36
countries (max of any messenger)
36
WhatsApp
28
Telegram
30
Signal
How we computed this

Filtered the platform-scores API to {signal, telegram, whatsapp} and reported the count of distinct countries with blocking evidence per messenger.

$ curl "https://api.voidly.ai/v1/platforms/scores"
GET /v1/platforms/scoresView platform risk scores →
F07Network#

38 active probes ran 24.5K measurements across 78 domains in the last cycle

Voidly operates a global probe network that tests a curated list of domains every 5 minutes. Across the most recent batch, 5% of attempts were blocked or interfered with. The network spans 18 countries and 5 detected blocking methods.

38
active nodes
24.5K
probes
78
domains tested
18
countries covered
5%
block rate
How we computed this

Live aggregate of the probe network. "Block rate" = blocked_count / total_probes from the last fetched probe stats payload.

$ curl "https://api.voidly.ai/v1/probe/stats"
GET /v1/probe/statsSee probe network →
F08Methodology#

Voidly probes detect 5 distinct blocking techniques in active deployment

Censorship infrastructure varies in technique. Voidly's probes classify each blocked attempt into one of: http-redirect, dns-poisoned, tcp-reset, tcp-timeout, tcp-refused. Method diversity matters because blocking techniques imply different mitigation strategies (e.g. encrypted DNS defeats DNS poisoning but not TCP reset).

5
methods detected
http-redirect
technique
dns-poisoned
technique
tcp-reset
technique
tcp-timeout
technique
tcp-refused
technique
How we computed this

The probe agent classifies each failed connection by signal: NXDOMAIN/poisoned-A → "dns-poisoned"; RST → "tcp-reset"; timeout → "tcp-timeout"; HTTP 30x with self-redirect → "http-redirect"; etc.

$ curl "https://api.voidly.ai/v1/probe/stats"
GET /v1/probe/statsRead methodology →
F09Growth#

VE leads cumulative incident count with 64 documented events

Voidly's top-5 most-active jurisdictions for documented censorship and disruption events. Rank reflects cumulative count across the live archive, not necessarily current severity.

64
VE incidents
64
VE
55
EG
55
IR
53
RU
52
NG
How we computed this

Top countries are returned directly by the stats endpoint, ordered by total incident count in the live database.

$ curl "https://api.voidly.ai/data/incidents/stats"
GET /data/incidents/statsView VE country page →
F10Censorship#

97.2% of all documented incidents are classified critical-severity

Voidly grades each incident on a three-tier severity scale (info / warning / critical) using a confidence-weighted rule combining blocking method, scope (number of services / ASNs), and corroborating sources.

97.2%
critical-severity share
2.6K
critical
75
warning
How we computed this

Sum of incidents grouped by severity = total. critical / total × 100 is the critical share. Warnings are typically transient probe-level interference; criticals span multiple measurements or sources.

$ curl "https://api.voidly.ai/data/incidents/stats"
GET /data/incidents/statsSee severity rubric →
F11Topics#

7 countries have severe or high censorship of women's health and group-rights content

Filtered to Citizen Lab category WMN/GRP, this topic isolates censorship of reproductive health and gender-related information. China ranks highest.

7
severe + high countries
100
China
67
Iran
67
Russia
How we computed this

Topic filter restricts the censorship index to evidence rows whose blocked domains map to the WMN or GRP Citizen Lab category. Country scores are then recomputed within that subset.

$ curl "https://api.voidly.ai/data/censorship-index.json?topic=women-health"
GET /data/censorship-index.json?topic=women-healthSee per-topic breakdown →
F12Topics#

Press freedom under threat: 3 countries severely restrict access to news websites

Filtered to Citizen Lab category NEWS, this topic shows where independent and international news outlets face network-level blocking. China leads with a topic-specific score of 100.

3
severe + high countries
100
China
65
Iran
61
Russia
How we computed this

Topic filter restricts the censorship index to evidence rows whose blocked domains carry the NEWS Citizen Lab category. Score is the same blend (block rate × breadth × method severity) computed within the subset.

$ curl "https://api.voidly.ai/data/censorship-index.json?topic=news"
GET /data/censorship-index.json?topic=newsSee per-topic breakdown →
F13Privacy#

8 countries actively block circumvention tools (VPNs, Tor, proxies)

When a regime moves to block circumvention tools — VPN providers, Tor bridges, or anti-censorship proxies — it signals a maturation of censorship infrastructure. China ranks highest in this topic.

8
severe + high countries
100
China
80
Iran
73
Egypt
How we computed this

Topic filter restricts to circumvention-related domains (vpn, tor, anonymizer categories). Score is recomputed against just those domains.

$ curl "https://api.voidly.ai/data/censorship-index.json?topic=circumvention"
GET /data/censorship-index.json?topic=circumventionSee per-topic breakdown →
F14Methodology#

Voidly aggregates 2.2B+ raw OONI measurements (aggregated source) into a single live API

The dataset behind these findings combines 312.8K live measurements over the last 30 days with a 1.6M historical record archive going back 2016-07-15. Updated every 6 hours.

312.8K
live measurements (30d window)
1.6M
historical records
130
countries
0.998 (internal 5-fold cross-validation)
model F1
How we computed this

Numbers are read directly from the metadata field of the live censorship-index payload. The 6-hour refresh comes from the cron job that regenerates ooni-data.json from the source DB.

$ curl "https://api.voidly.ai/data/censorship-index.json"
GET /data/censorship-index.json (metadata field)See methodology →

Data overview

New incidents in our recent sample

Counted from the most recent 200 incidents returned by /data/incidents. Pipeline activity, not absolute volume.

26-05
200

Evidence sources

Voidly cross-checks signals from multiple independent measurement networks.

ooni45,537 (53.3%)
ioda21,626 (25.3%)
censoredplanet17,241 (20.2%)
Voidly-Community507 (0.6%)
Voidly478 (0.6%)

Reproduce any finding

Each finding above can be reproduced with a single curl. The full pipeline — ingestion, scoring, storage — is documented and the underlying data is open. Use the “How we computed this” expander on each card for the exact API call.

# All incidents (paginated, JSON)
curl "https://api.voidly.ai/data/incidents?limit=200"
# Stats summary (totals, growth, by-severity)
curl "https://api.voidly.ai/data/incidents/stats"
# 7-day shutdown forecast (Voidly Sentinel)
curl "https://api.voidly.ai/v1/sentinel/global_heatmap?min_risk=0"
API docs →·Methodology →·Open data hub →·RSS feed →·HuggingFace dataset →
Press / journalists

Need a tailored finding for a story?

We can co-author or vet a story for you, run a custom query against the dataset, or pull a deeper drill-down for a specific country, ISP or platform. We're fast and we don't require attribution agreements.

Email press@voidly.aiFor journalistsPress kit

All findings released under CC BY 4.0 · cite as voidly.ai/findings · last refresh Thu, 21 May 2026 04:12:33 GMT

See also

Cite this page

All Voidly Research data is licensed under CC BY 4.0. You can redistribute, remix, and build on it freely with attribution.