AI providers should disclose how workloads survive data-center outages

The accounts do not settle every detail about the incidents. The Record reported that Yandex acknowledged a power outage at Sasovo but did not publicly attribute that outage to a drone strike. Russian authorities said air defenses intercepted Ukrainian drones over several regions, including Ryazan, while local authorities did not report damage. Those statements do not independently establish what caused the damage Yandex reported or what the intended targets were.

The service impact is clearer. DatacenterDynamics reported that Yandex’s status page showed cloud zones unavailable and the platform operating in a limited mode, with at least 84 services taken down during the incident. That is a reported count of services taken down, not evidence that they were permanently lost.

The Record cited internet-monitoring group NetBlocks’ observation that Yandex connectivity fell to around 70% of normal during the Sasovo disruption. That figure concerns Yandex connectivity, not the Russian internet as a whole or the performance of every AI service.

Workload transfers do not answer every question

Yandex’s response is central to any assessment of resilience. The Record reported that the company said it moved workloads to other facilities, which were operating normally, though restrictions remained. That describes a continuity response, but it is not an independent evaluation of how well transfers worked or how much capacity was available.

Yandex also warned, “The ability to create new resources in our cloud infrastructure may currently be limited,” The Record reported. That qualification matters: moving workloads elsewhere does not by itself show whether every service continued, whether some tasks were delayed or whether the receiving sites had enough capacity to take on more work.

The reports do not identify which workloads moved or whether any were used for AI development. Reuters reported that two of Yandex’s three supercomputers used to develop its AI model were housed at Sasovo. Yandex has not said whether those two machines were damaged, and the site’s shutdown alone does not establish that AI model development stopped.

Yandex said it was still assessing its infrastructure and could not say when equipment at the affected data centers would be running again. The available reporting also does not quantify how many users or businesses were affected, how long disruptions lasted or which customer workloads remained available through transfers.

That uncertainty cuts both ways. A facility shutdown does not prove that a provider’s services have broadly failed; a report of workload transfers does not prove that its continuity arrangements were sufficient. Customers need information about what stayed available, what was interrupted and how long recovery took to judge the service they depend on.

Providers should report continuity without exposing defenses

There is a real limit to what operators should disclose during or after an attack. Detailed facility layouts, security measures or backup routes could expose sensitive information. But providers can report service availability and recovery performance without publishing a blueprint for defending their sites.

We recommend that major cloud and AI providers publish clear post-incident accounts once the immediate response allows it. Those reports should identify which services were disrupted and restored, whether workloads moved to other facilities, how long recovery took, and what capacity constraints or unresolved dependencies remain. Where an independent technical assessment is available, providers should share its findings while withholding sensitive security details.

The concentration of two of Yandex’s three AI-development supercomputers at Sasovo makes the unanswered equipment question consequential, not settled. The reports justify scrutiny of continuity planning; they do not support treating a wider failure of AI-dependent services as established.