What happened

The first reports appeared on the night of August 28. Media outlets cite similar times: around midnight according to outage-monitoring services, around 12:45 AM Moscow time, around 1:30 AM according to other publications.

Everyone described the same symptom: the page would start loading and then cut off. “The page keeps loading endlessly, and then the connection drops,” Rambler writes. Through circumvention services and from foreign addresses, the site worked normally, as confirmed by both a Business Online correspondent and users who checked from foreign IPs.

The number of complaints varies between publications, and the reason lies in the sources. Some outlets cite the Detector404 service, which counted 219 reports over 24 hours, 98% of them from Russia. Others cite Downdetector.su. Others cite data provided by TASS: 84% of complaints about the site, 7% about the app, and 7% about a general outage. The services differ, and so do their methodologies, so you can’t compare their numbers directly. In March 2026, Roskomnadzor (Russia’s communications regulator) warned that Detector404 could be used to spread fake news.

Our testing methodology

The idea is simple. If a resource is unavailable to some users, that should show up from four directions at once: in traffic on the resource’s own side, in active measurements from independent observers, in the state of the networks, and in how people themselves experience it. If all four point the same way, it is no coincidence.

Every measurement needs a control. Without a control group, you can explain away any graph however you like.

Wikipedia’s statistics

The Wikimedia Foundation publishes hourly traffic for its language editions with about a three-hour delay. This is a direct source — it shows not complaints, but how many people actually opened pages.

We pulled data for the Russian edition for the night of the event and compared it with the same hours the previous day.

Time (Moscow)	Before	After	Change
03:00	388,730	214,437	-44.8%
04:00	296,507	188,409	-36.5%
05:00	308,701	188,121	-39.1%
06:00	372,136	215,942	-42.0%
07:00	492,001	283,330	-42.4%
08:00	607,898	344,462	-43.3%
09:00	712,507	662,593	-7.0%

The average drop over these six hours came to 41%.

The deviation from normal starts at 1:00 AM, when traffic drops 36% compared to the same hour the day before, and reaches 55% by 2:00 AM. That’s exactly when the first complaints came in. By 9:00 AM, the gap narrows to 7%, meaning traffic had almost returned to normal.

The control experiment

A drop by itself proves nothing. After all, the whole service could just go down entirely, right?

So we looked at the English edition for the same hours. It showed changes ranging from -2.3% to -6.5%, averaging -5%. That’s normal day-to-day fluctuation so that we can rule out a service-side outage.

Independent measurements

The OONI platform (Open Observatory of Network Interference) runs active availability checks through volunteers. There’s a nuance here: if you query data for the wikipedia.org domain, there’s almost nothing there, only eight measurements, zero anomalies. The signal shows up on the ru.wikipedia.org domain, which is where the Russian edition actually lives. Outlets that wrote about anomalies in OONI looked at encyclopedia’sia’s general address, which is why their picture came out fainter.

Here’s the picture for the Russian edition. From August 18 to 26 — two anomalies out of 282 measurements, or 0.7%. On August 27 — 11.4%. On August 28 — 30.8%. That’s a forty-fold increase over the baseline.

OONI’s confirmed-blocking field reads zero on every single day. The project applies that label when it sees a block page with a notice explaining the ban. No block page appeared; the connection simply never got through.

We checked individual operators for August 27 and 28. Some networks showed anomalies in three of four measurements and four of twelve. Others had zero anomalies over the same two days.

Neither a service-side outage nor a general network failure behaves this way. Either one would hit everyone equally. Operator-level selectivity suggests a configuration applied unevenly, not everywhere at once. We described this exact pattern in our piece on the Apple outage: some networks affected, others clean at the very same hour. It looks like a partial block or a test of restrictions.

Three measurements line up

The outlet Bumaga ran a reader poll. Of 61 respondents, 32 noticed no problems, 24 couldn’t open the site that night, and five still couldn’t afterwards. That works out to 47% affected. The sample looks tiny and unrepresentative, and you shouldn’t rely on it alone. But now let’s line it up against the other data. OONI — 31% anomalous measurements. Traffic — 41% anomalous measurements. The poll — 47% affected.

Different methods, different samples, different kinds of data, and yet the result falls into one range: somewhere between a third and a half. The restriction wasn’t total. Roughly half the users noticed nothing at all, which fits with operator selectivity.

Different statements

We don’t know where government agencies stand on this particular event; they haven’t commented. The most recent statement dates back to January 2026, when Roskomnadzor told TASS that no requests to block Wikipedia had come in. That said, officials have long talked about the possibility of blocking Wikipedia.

Wikipedia remains one of the few large international resources still operating in Russia without restrictions. Pressure on it has built up for years: in 2024, courts fined the Wikimedia Foundation 9 million rubles (about $119,700), then another 3.5 million rubles (about $46,550) in 2025. By May 2025, 232 pieces of content remained unremoved, and Roskomnadzor said it had “enough tools to motivate the company.”

For instance, Anton Gorelkin, deputy chair of the State Duma’s information policy committee, explained back in 2025 why the encyclopedia still hadn’t been shut down. He said legal grounds exist, and “what’s left is to make sure this doesn’t cause users any sharp discomfort.” He went on to explain the mechanics: valuable information has already been duplicated on the Great Russian Encyclopedia, Ruwiki, and Runiversalis portals, and all that’s left is to make sure that a link to a Wikipedia article offers the user its safe version on a Russian equivalent instead.

One of these equivalents, Ruwiki, launched in the summer of 2023, copying 1.9 million articles from Wikipedia itself at launch, some of which were later edited. The project gets promoted heavily, right down to a themed train car on the Moscow metro. There’s also an academic assessment: a paper by Vestel and Degaetano-Ortlieb, presented at the LaTeCH-CLfL workshop in 2026, compares the fork’s language with the original and examines the linguistic traces of war propaganda.

Runiversalis, at the domain runi.rf, and the Great Russian Encyclopedia (launched in 2019 with support from Mintsifry, Russia’s Ministry of Digital Development) also used to exist. The encyclopedia has since shut down, and its articles went to Ruwiki.

Sergei Vilyanov, an analyst at the Fintech Lab accelerator, believes that “blocking the whole site isn’t just illogical, it’s simply pointless,” and notes that Wikipedia effectively has no full-fledged equivalent, not even at the level Russian replacements for YouTube have reached.

Five theories we checked

Theory one: global Wikipedia outage—is proven. The Wikimedia Foundation’s official status page for August 28 reported no incidents, same for August 27 and 26. The platform’s metrics were normal. The page did honestly record real outages. On August 25, it logged a global outage with sites unreachable. Three days later, on the night of the event, that page was empty.

Theory two: a mass internet shutdown in Russia—is proven. Cloudflare Radar shows Russia’s traffic for the week following its usual daily rhythm, with no drop on the night of the 28th. IODA (a platform run out of Georgia Tech, in the US state of Georgia, that independently tracks network availability) holds steady at 100% over the same window, both for the visibility of Russian networks in global routing and for active probing of address blocks. Networks didn’t disappear, and routes weren’t withdrawn.

Theory three: Wikipedia went down together with the banks. This claim showed up in some retellings, but it has also been disproven. We checked separately, and there were no reports of bank outages on the night of August 28. Every piece of coverage we found about bank outages referred to other dates, ranging from April to August 18. The phrase “as previously reported” in one article was pointing back to an earlier event.

Theory four: this happens regularly; people just don’t notice —disproven. In Yandex’s search statistics, we came across queries with specific past dates: “Wikipedia not working July 30” and “August 19.” We checked daily traffic over two months, from July 1 to August 27. The median was 18.12 million views per day, with normal fluctuations around 3.5%. July 30 showed -2.4%; August 19 showed +2.6%. No date stands out. August 28 is the only case like this in these two months.

Theory five: a selective network restriction. The only one that fits all the data: a drop limited to the Russian edition, normal operation of the English one, no incident at the Wikimedia Foundation, intact networks, selectivity by operator, no block page, and everything working through circumvention.

We don’t have a direct proof. There’s no official confirmation either. But none of the other theories fit.

What we don’t know

• The mechanism. The description “the page started loading and cut off” resembles how deep packet inspection equipment behaves, but we don’t have measurements to pin down the exact stage of the cutoff so that we won’t claim that.
• The reason. No official explanation has come from Roskomnadzor, the Wikimedia Foundation, or internet providers. In the Apple case, at least one theory was that it was the authorities’ response to Russian apps being removed from the App Store. Wikipedia, while it has long irritated officials, hadn’t given any obvious fresh reason for a block.
• The geography. The four available data slices diverge enough that the only thing we can say with confidence is this: different regions were affected, but not all of them.

You can check it yourself.

Everything we did here is available to anyone.

Hourly traffic for the Russian edition (change the dates at the end of the URL): https://wikimedia.org/api/rest_v1/metrics/pageviews/aggregate/ru.wikipedia/all-access/user/hourly/2026082600/2026082823

Same for English, as a control: replace ru.wikipedia with en.wikipedia.

OONI data in machine-readable form: https://api.ooni.io/api/v1/aggregation?probe_cc=RU&domain=ru.wikipedia.org&since=2026-08-18&until=2026-08-29&axis_x=measurement_start_day

Breakdown by operator: add to the end:&axis_y=probe_asn

Wikimedia platform status: https://www.wikimediastatus.net/

Availability of Russian networks: https://ioda.inetintel.cc.gatech.edu/country/RU

Our own access-mode checking tool: csrlab.org

Conclusion

Established: the drop happened and is measurable; it hit the Russian edition, not the English one; it didn’t originate on Wikipedia’s network; shutdowns didn’t accompany it; it affected roughly half of users; and it distributed unevenly across operators.

Not established: exactly who did this and why.

We’re continuing to monitor this and will update the piece if official explanations or new episodes emerge.

Sources and methodology

Traffic data came from the Wikimedia Foundation’s open statistics interface, licensed under CC0. Data on availability measurements came through OONI’s open interface. Network status came from Cloudflare Radar and IODA (Georgia Tech, in Georgia, US). Bumaga published the reader poll. The channel ZaTelecom raised a sceptical view of whether the OONI data was sufficient, and that’s a fair point: the sample is small, around three dozen measurements per day. We took this into account and relied primarily on traffic statistics with millions of observations.

Sources

Our own measurements

• Hourly and daily traffic for Wikipedia’s language editions, via the Wikimedia Foundation’s open statistics interface. Our main source: the drop in the Russian edition and the control comparison with English. Data licensed under CC0.
• OONI availability measurements, queried directly. The rise in anomalies from a 0.7% baseline to 30.8%, and the breakdown by autonomous system.
• Wikimedia platform status. No incidents recorded on August 26, 27, or 28, against a logged outage on August 25.
• Cloudflare Radar for Russia. The country’s traffic showed no drop on the night of the event.
• IODA, Georgia Tech (in the US state of Georgia). Network visibility in routing and address-block availability.
• Google Trends and Yandex Wordstat. Search demand and its baseline levels.

Reports on the event

• RIA Novosti, on the outage according to Detector404. Tal source for the figure of 219 complaints in 24 hours and the 98%-from-Russia share.
• Rambler, on the outage according to Downdetector.su. The start time and another geographic breakdown.
• Rambler, “Wikipedia hit by outages in Russia”. A description of the symptom.
• Business Online, “Wikipedia won’t open for Russians”. A correspondent’s own check, circumvention services working, and the complaint breakdown.
• Argumenty i Fakty, on outages starting at midnight. The start time.
• Ridus, citing Detector404. A repeat of the data on the share of reports from Russia.
• RZN.info, complaints in Ryazan. Another geographic breakdown.
• Bumaga, reader poll. A self-selected sample, 61 votes.
• ZaTelecom, on whether the OONI data is sufficient. A sceptical position that we factored in.

Context and background

• News.ru, “Just copy-paste it”: why Wikipedia isn’t blocked in Rus—Antonton Gorelkin’s quote.
• Habr, on Roskomnadzor’s ignored requirements. Sergei Vilyanov’s quote.
• Fontanka, Roskomnadzor’s comment to TASS, January 2026.
• RIA Novosti, on the fines against the Wikimedia Foundation.
• Hi-Tech Mail, on the unremoved content.
• Vestel S., Degaetano-Ortlieb S. Modelling Linguistic Imprints of War Propaganda in a Russian Wikipedia Fork. LaTeCH-CLfL workshop, 2026. An academic comparison of Ruwiki’s language with the original.

Our own pieces

• Our piece on the Apple and App Store outage. It describes the same method and the same pattern of operator selectivity.
• The access-mode web checker. A tool you can use to check your own network.