Fact check: Ukraine hit 3 Yandex data centers Oct 8-11 (Sasovo, Kaluga, Vladimir). Yandex confirmed it; Reuters and The Register reported it.

Damaged Yandex data center

But the claim Russian bot networks “went offline” is unproven. It rests on a misread Cloudflare chart of crawler traffic, not social media accounts. No platform announced takedowns.

The strikes: confirmed

The strikes hit three sites in four days, and Yandex itself confirmed the damage.

In Sasovo, in the Ryazan region, a drone attack overnight Oct 7-8 caused a fire and halted operations completely. The site housed two of Yandex’s three AI supercomputers, according to Reuters, citing Yandex’s own disclosures.

In the Kaluga region on Oct 9, several modules were completely knocked out and operations partly shut down. Yandex confirmed the hit. Ukrainian drone maker Firepoint said its FP-1 one-way attack drones were used, according to Reuters on Oct 9.

In Vladimir, overnight Oct 10-11, operations were completely suspended and Yandex Cloud went into emergency mode. Regional governor Aleksandr Avdeyev confirmed a drone hit on an infrastructure facility plus a fire. Two substations went offline and about 5,000 residents lost power.

More than 80 services were disrupted, including Yandex Cloud, Yandex Pay, taxi, and the Alice AI assistant. NetBlocks measured Yandex Cloud connectivity at about 30 percent of normal. Ukraine frames the strikes as retaliation for Russian strikes on Ukrainian data centers since late September.

Where the bot claim came from, and why the chart does not prove it

The earliest traced version of the bot claim is an Oct 8, 2026 post on X by security researcher Jeremiah Fowler, who posted a Cloudflare chart and read it as a Russian bot takedown. It spread on Oct 11 through high-reach aggregator Mario Nawfal, journalist Jamie Dupree, and a misleadingly headlined Tom’s Hardware story, which was later softened.

The chart itself measures something else. Cloudflare Radar data for Yandex’s network (AS13238) showed automated traffic falling from about 57 percent of the two-week peak after the first strike to about 0.4 percent after the third. In Cloudflare’s terms, “bot traffic” means automated non-human HTTP requests: search crawlers, indexers, ad-fraud checks, and scrapers. Yandex’s own search crawler is itself a Cloudflare-verified bot, and about 99.6 percent automated traffic is normal for a data center network. The chart therefore shows Yandex’s own infrastructure going offline, not social media accounts going offline.

No platform announced a Russian bot takedown in this window: no announcements from X, Telegram, Meta, or YouTube were found. No disinformation-monitoring source corroborated a drop in bot activity, according to a canvass by Joohn Choe on Oct 12 covering DFRLab, NATO, EU disinformation resources, NewsGuard, CheckFirst, Alliance4Europe, AFP, Reuters, and Snopes.

President Volodymyr Zelenskyy said at the Warsaw Security Forum on Oct 12 that social media became “easier to breathe” with “fewer bots” after the strikes. That is an interested party’s observation, not a measurement. His office later told BBC Verify that he meant the atmosphere on social media, not a measured fall in bots.

One real takedown in the same window is unrelated to the strikes. On Oct 8, OpenAI banned ChatGPT accounts tied to Russian and Iranian influence operations, according to AFP reporting. That was a separate, routine enforcement action against ChatGPT accounts, not social media accounts, and it is not evidence that the strikes shut anything down.

The harder claim: engagement dropped, proving the strikes killed Russian influence

A follow-up check on Oct 14 tested a harder version of the claim: that engagement on Russian influence, pro-Russian, or right-wing accounts measurably dropped after the strikes. That version is false. The specific numbers circulating are fabricated, cherry-picked, or misread.

The most widely shared number was fabricated. A chart posted on Oct 11 by Czech MEP Tomáš Zdechovský showed AfD online activity falling 30 percent and then 56 percent, with no platform, source, metric, or account list given. The chart was traced to a Threads user, its x-axis ran to Oct 15, days after it was posted, and its author apologised on Oct 12, saying it was AI-generated and unverified. It was flagged by Lead Stories, Mimikama, Capital, and Rohtext. Polish analyst Jarosław Wolski recaptioned the same chart and then deleted his post and apologised.

Another viral number was cherry-picked. A chart comparing Tucker Carlson’s average views claimed a 71 percent drop, but it compared 7 posts with 4 posts, and its own footnote admitted the newer posts had had less time to accrue views. That is an accrual bias, not a measurement of a bot outage.

The strongest independent test points the other way. Graphika, publishing Oct 13, tested AfD, MAGA, and Reform UK communities against live community maps. In the AfD-related community on Sunday, Oct 11, it counted 40,365 posts across 1,022 accounts, against 42,040 posts and 1,037 accounts one week earlier: a 4 percent fall, well within the normal range, since prior Sundays had swung by up to 38 percent. MAGA mentions on X rose 25 percent in the first 8 hours after the Vladimir strike, compared with the same hours on the prior four Sundays. Graphika’s conclusion was plain: none of the communities named in the claims lost their active accounts or heaviest posters.

Other trackers reached the same null result. Common Consultancy, measuring Facebook comments, found AfD between 8 percent down and 5 percent up, and MAGA between 4 percent and 10 percent up: ordinary fluctuation. Polisphere, CeMAS, NASK in Poland, dTeam in the United States, antibot4navalny, and CAT-UA in Ukraine found no significant decline and no loss of accounts tied to the strikes. No DFRLab, NewsGuard, EU DisinfoLab, NATO StratCom, Recorded Future, or platform transparency report was found showing a drop.

There is one narrow, partial exception, and it does not rescue the claim. Ukraine’s Center for Countering Disinformation, in preliminary findings, said there was no significant overall decline in Russian bot-network activity, but that volume and frequency in some Pravda, InfoDefense, and Nod of Time clusters fell by about half for about 12 hours, then resumed. That is a single government-stakeholder source, preliminary, with no dataset or method published. At most it supports a brief, partial disruption of some automated dissemination, not a collapse of influence.

A calendar confounder also explains part of the appearance of a drop. Oct 11, 2026 was a Sunday. Graphika found that mentions of all major German parties dipped that Sunday morning by similar amounts. A drop shared by rivals points to the calendar, not to targeted disruption.

Why the mechanism does not work as a general explanation

Russian influence operations generally do not run on Yandex. Darren Linvill of Clemson’s Media Forensics Hub told Snopes that the Russian propaganda campaigns he studies do not use physical infrastructure inside Russia; they outsource to services run outside Russia. Cybernews reported that, as a rule, Russian troll farms do not run bots directly through Yandex, since platforms usually block traffic from known Russian corporate data centers, so operators use residential proxy networks, compromised devices, commercial VPNs, and Western cloud hosting. Yandex told Cybernews that its Cloud rules ban mass manipulation and spam and that violators are blocked.

Operation-by-operation checks reach the same conclusion. Doppelganger used European and United States hosting plus Aeza bulletproof hosting. Storm-1516 used United States hosting and growing Cloudflare use. Pravda network domains resolve to Cloudflare. Meliorator domains were bought from an Arizona registrar, with no Yandex link. Of 75 AfD-related domains checked, none relied on Yandex.

Verdict

The Oct 12 verdict was partly true: the strikes are confirmed, and the claim that Russian bot networks were shut down is unproven. The Oct 14 verdict on the harder claim is false: the strikes did not kill Russian influence, as proven by reduced engagement, because no such engagement collapse is measured. The only documented effect on influence operations is a brief, partial, single-source disruption of some automated clusters for about 12 hours.

What would change this verdict: platform data showing a sustained, statistically unusual fall in posting or engagement for a pre-registered set of attributed Russian influence accounts, starting at a strike and absent in control groups, plus hosting evidence tying those accounts to the specific Yandex facilities that were hit.

Sources