top of page

Yandex Cloud Data Loss: 2.84% Unrecoverable After Strikes

43 minutes ago
11 min read

Yandex says it cannot recover 2.84% of affected Yandex Cloud customer data following Ukrainian drone strikes against its data centers. The Yandex Cloud data loss disclosure accompanies a recovery figure that initially sounds reassuring: engineers restored access to 97.16%. For customers whose records remain missing, however, the announcement establishes a permanent loss rather than another delay.

The company’s November 10 update also changes the starting conditions for customers creating virtual machines, software-based computers running on shared physical servers. Backups now begin enabled, although customers can disable them. Yandex is also expanding its infrastructure, introducing another availability zone, a separate infrastructure location intended to help isolate failures.

The tension is between restoring a cloud platform and restoring the businesses that depend on it. AWS makes a comparable distinction through its division of provider and customer responsibilities. Yandex’s disclosure brings that boundary into focus: infrastructure can return while customer information remains unavailable forever.

What the Yandex Cloud data loss figures actually measure

The recovery percentage describes affected cloud customer data, not all information held across Yandex’s businesses.

That distinction changes the meaning of the headline. “Russia’s Google” is a familiar shorthand for Yandex, but it obscures which service suffered the disclosed loss. The announcement concerns Yandex Cloud, rather than a measured percentage of the company’s search index, consumer accounts, or global information holdings.

Yandex’s customer data recovery statement attributes the unrecoverable portion to critical equipment damage and missing customer backups on its platform. It also says customers using its backup tools preserved all their data. Those are company statements, rather than findings from an independent assessment.

The percentages do not establish how many customers lost information. A share of data is different from a share of accounts, organizations, applications, or transactions. Converting the disclosed figure into a customer count would require information that the announcement does not provide.

The same caution applies to business impact. Lost storage capacity does not translate directly into lost revenue, interrupted transactions, or missing documents. Different records carry different operational value, and the announcement offers no breakdown that would support those calculations.

A missing archive and an incomplete transaction database can create very different recovery problems. The archive might matter primarily for historical access. The database might determine whether a company can reconcile orders, payments, and obligations.

These are illustrative consequences, not identified losses at particular Yandex customers. Their relevance is that a platform-wide percentage cannot describe an individual organization’s position. An apparently small unrecoverable share can still contain information essential to a particular business.

The timeline also needs care. The recovery disclosure appeared on November 10, while Tom’s Hardware published its coverage on November 12. Neither date should be presented as the date of a newly reported attack.

Interfax places the damaging attacks between October 8 and October 11. Its account of the October drone attacks connects the subsequent recovery effort to that earlier incident. The November development is the recovery assessment and accompanying policy changes.

That makes this more consequential than another service interruption update. An outage leaves open the prospect that unavailable records will return. A declaration that information cannot be restored changes what affected customers must plan around.

It also limits what readers should infer from the recovery headline. Restored access is meaningful progress, but the announcement does not establish that every affected application has resumed normal operation. Customer systems require more than readable storage to function.

Customers face a different recovery problem than their provider

A provider can restore infrastructure while customers still face incomplete records, disrupted processes, and unresolved business decisions.

Yandex’s immediate task involves recovering data and making computing resources available. Its customers face another task: determining whether their applications and records remain usable. The difference explains why a recovery percentage cannot serve as a complete account of the incident.

Consider an illustrative retailer whose application server returns with an incomplete order history. The website might accept new purchases while staff investigate earlier transactions. A functioning storefront would not establish that customer balances, deliveries, and refunds remained accurate.

For that business, recovery depends on reconciling information across systems. Payment confirmations, delivery records, customer messages, and accounting entries might help establish what happened. Whether those sources exist independently would determine how much reconstruction is possible.

A software team faces a related distinction. Replacing an application server might be straightforward if its code and deployment instructions survive elsewhere. Reconstructing missing customer information is a different problem, especially when no authoritative copy remains.

These examples explain who comes under pressure after a permanent-loss disclosure. Developers must establish what can run, while business owners must establish what can be trusted. Purchasing teams must determine whether the service’s actual protections match their original assumptions.

The Yandex Cloud data loss announcement therefore raises questions beyond whether servers are available again. Customers need to understand which records survived, when those records were last complete, and which business processes depend on them. Different answers can produce different reopening decisions.

The pressure also reaches internal reporting. Executives need a distinction between restored services and unresolved information gaps. Otherwise, a dashboard showing operational systems can conceal continuing work inside finance, customer support, or compliance teams.

There is a documentation dimension as well. Recovery decisions depend on knowing which systems hold authoritative information and who controls the copies. A searchable knowledge base can help teams locate those decisions, ownership records, and recovery instructions.

That organizational benefit has a clear limit. Searchable documentation cannot recreate missing production records. Its value lies in making responsibilities and dependencies easier to establish when an incident forces decisions.

Yandex itself faces pressure to describe recovery in terms customers can apply. Aggregate progress communicates scale, but affected organizations need account-specific answers. The remaining information gaps matter most where customers must decide whether to rebuild, reconcile, or retire a damaged system.

This creates a communication challenge after the technical recovery effort. The provider’s progress and the customer’s recovery can move at different speeds. Clear reporting needs to acknowledge both without treating either as the complete story.

The practical consequence is that permanent loss shifts attention toward business evidence. Teams need confidence in records and decisions, alongside confidence that infrastructure is functioning. That is the standard against which the recovery announcement ultimately matters.

Default backups change the starting point for new workloads

Yandex’s backup change reduces reliance on customers remembering to enable protection during creation of a new virtual machine.

Previously, customers had to activate backups themselves, according to the company. The updated default reverses that initial choice. Customers can still turn the feature off, so describing the change as mandatory protection would overstate what Yandex announced.

That distinction matters because defaults influence what happens when a customer makes no deliberate decision. Enabling protection at creation removes an initial omission from the setup process. It does not establish that the resulting configuration matches every application’s needs.

Nor does the announcement establish that existing virtual machines have all received equivalent protection. The stated change concerns newly created machines. Extending that claim to every existing workload would require additional evidence.

The strongest case for the change is straightforward. A customer who overlooks the backup setting starts from a more protective position. Yandex’s response acknowledges that the previous setup left an important choice dependent on customer action.

The skeptical interpretation concerns scope. A backup feature can be enabled without answering every question about recovery. Customers still need to understand what information it captures, how current that information remains, and what restoration requires.

These are boundaries of the announcement, rather than evidence that the new setting fails. Yandex has described a change in configuration. It has not published evidence here demonstrating successful restoration of every customer application.

The company’s statement that backup users preserved their data supports its case for wider adoption. However, that statement remains attributable to Yandex. It does not independently establish how every protected workload performed during the disruption.

The competing interpretations deserve equal care. Recovering most affected information after physical destruction represents a substantial recovery effort. Declaring some information permanently unrecoverable also identifies a protection gap with lasting consequences.

Neither observation determines whether customers understood the earlier arrangements. The public disclosure does not establish what each customer was told, which settings they selected, or which contractual protections applied. Those questions require more specific evidence.

The broader comparison is familiar in cloud computing. AWS’s shared responsibility model assigns infrastructure responsibilities to the provider and configuration responsibilities according to the services customers select. For its virtual machines, customers manage important aspects of application resilience.

AWS also describes backup, versioning, and replication as customer considerations for data resilience. Versioning preserves earlier versions of information; replication maintains additional copies. These mechanisms address different recovery needs and require decisions about how they are used.

The comparison provides context, rather than proof that another provider would withstand the same incident. Services differ, and no equivalent failure test is established here. The relevant similarity is the need to understand where platform protection ends and application responsibility begins.

Yandex Cloud backups now start from a different default for new machines. The resulting benefit depends on what customers retain, configure, and successfully recover. That is a narrower claim than universal protection, but still a meaningful change.

New availability zones expand options, with limits

Additional infrastructure gives customers more deployment choices, but its value depends on the services and recovery arrangements available there.

Yandex’s November 10 blog post announced a November 11 launch for the availability zone identified as ru-central1-f. The same update documented launch restrictions. Its availability zone expansion therefore describes an additional option with qualifications, rather than an unrestricted replacement for every existing deployment.

The documented restrictions include resource quotas, which limit the amount of computing capacity customers can obtain. The post also describes service-specific limitations at launch. Some private connection traffic would initially receive processing through other zones.

Those details matter because an application can depend on several services together. Moving its computing component does not establish that every supporting component has equivalent independence. Customers need an application-level picture of what the new location supports.

For example, an illustrative business might run its application in the new zone while relying on a database elsewhere. That arrangement can still be useful. Its recovery behavior depends on the relationship between those components, rather than the application server’s location alone.

This is why the phrase “secret availability zones” needs restraint. Withholding precise facility locations does not itself demonstrate protection from further attacks. The operational questions concern separation, dependencies, available capacity, and whether customers can restore functioning systems.

Physical secrecy and technical independence address different concerns. A less publicly visible location might still depend on shared services. Conversely, useful information about recovery capabilities can be published without exposing a facility’s precise address.

The analytical issue is whether expansion creates practical recovery options. Spare computing capacity matters when customers can obtain it and connect it to surviving data. Capacity that requires unavailable dependencies offers a more limited benefit.

Migration also deserves careful language. Moving a workload means transferring or rebuilding its operating components in another location. It does not necessarily produce an independent historical copy of its information.

Likewise, replication and backup should not be treated as interchangeable guarantees. A replica supports continued access to copied information. A backup preserves information for later restoration, subject to its configuration and condition.

The distinction matters when a business needs to recover an earlier valid state. Additional live copies and retained historical copies serve related purposes, but they answer different failure scenarios. A public announcement needs enough detail for customers to understand which protection it describes.

None of this establishes that Yandex’s new zone is ineffective. The published restrictions instead identify where customers need more information. Improvements to service coverage and capacity would make the expansion more useful for a wider range of applications.

For Yandex Cloud backups, the location of recoverable information is part of the same question. Protection becomes operationally valuable when surviving copies can support restoration somewhere usable. New infrastructure and backup defaults need to work together to improve that outcome.

The disclosure leaves responsibility and impact unresolved

The company’s explanation identifies damaged equipment and missing platform backups, but it does not settle every question about responsibility.

Tom’s Hardware emphasizes the contrast between the recovery headline and the unrecoverable remainder. Its discussion of the backup policy change also highlights how Yandex frames customers’ backup choices. That framing deserves scrutiny without assuming that it proves customer fault.

An absent platform backup does not establish that a customer kept no copy anywhere else. Some organizations might maintain independent records or exports. The announcement does not provide enough detail to determine which affected customers have those alternatives.

It also does not establish that every missing record is recoverable through another route. External copies might be absent, incomplete, or too old for particular business needs. Both possibilities remain open without customer-specific evidence.

The Yandex Cloud data loss figure therefore describes what the provider says it cannot restore. It does not measure the final unrecoverable business loss across all affected organizations. Establishing that total would require evidence about customers’ other copies and reconstruction efforts.

Another unresolved issue concerns the measurement itself. The public statements do not provide a detailed methodology for interpreting the percentages across different customer services. Readers should resist treating them as a complete technical assessment of every affected system.

There is also no basis here for turning the percentage into a precise volume of destroyed information. Such a calculation would need a verified denominator covering the same affected population. Broader historical storage figures would not automatically provide it.

Responsibility is similarly specific. Infrastructure design, product defaults, customer configuration, and contractual commitments can all matter. Their relevance depends on the service and the circumstances, rather than a general assumption that cloud storage includes every desired protection.

A provider’s reference to shared responsibility does not resolve whether its explanations were clear. A customer’s reliance on cloud services does not establish that every recovery feature was included. The evidence needed to assess those questions goes beyond the brief recovery announcement.

The missing independent assessment is especially important for interpreting broad claims. Company statements establish what Yandex reports about its recovery work. They do not provide outside validation of the data-preservation claim or the completeness of each restoration.

That limitation should shape the language of coverage. “Yandex says” accurately describes the source of the figures. Declaring the entire platform fully recovered would go beyond what the disclosed permanent loss and service qualifications support.

The same restraint applies to geopolitical conclusions. The incident demonstrates that physical damage affected a major cloud provider’s customers. It does not establish that all cloud services share identical vulnerabilities or that migration to a particular competitor would eliminate them.

The most defensible judgment concerns the gap between service recovery and information recovery. Yandex’s announcement makes that gap explicit. How it affected individual customers, and how responsibility is allocated, remains less fully documented.

What would show that recovery protections have improved

The next meaningful evidence will concern recovery results, protection of existing workloads, and usable capacity in the expanded infrastructure.

The first signal to watch over the coming months is more detailed recovery reporting. Evidence describing what customers successfully restored would strengthen confidence beyond an aggregate percentage. An independent assessment would add further weight, if one becomes available.

Useful reporting would distinguish access to surviving information from restoration of functioning applications. Those outcomes answer different customer questions. Greater detail would make it easier to assess whether the recovery process met the needs of affected businesses.

The second signal is clearer coverage of existing workloads. The announced default concerns new virtual machines. Further disclosures about older machines would show how broadly the revised approach extends across the installed customer base.

That signal matters because new deployment settings change protection gradually. Existing applications remain relevant to the provider’s overall recovery position. Evidence of wider backup coverage and successful restoration would strengthen the case that the policy change addresses continuing exposure.

The third signal is progress against the new zone’s launch limitations. Broader service availability, accessible capacity, and clearer dependency information would improve its usefulness. Continued restrictions would narrow what customers can reasonably expect from the expansion.

These signals should be evaluated together. Better backup coverage helps preserve recoverable information. Available infrastructure gives customers somewhere to use it. Successful restoration connects those improvements to the business outcome that matters.

There is no announced deadline here guaranteeing an independent assessment or complete removal of the launch restrictions. These are observation points, rather than promised milestones. Their value lies in making future progress measurable without inventing a forecast.

For buyers outside Yandex’s customer base, the incident creates a concrete comparison point. They can examine what their own providers disclose about recoverable records, deployment choices, and application dependencies. The relevant question is how those arrangements perform under disruption.

For developers, the strongest evidence will come from the systems they actually operate. A general provider statement cannot establish whether a particular application has current, usable recovery copies. That remains an application-specific question even when a platform offers extensive protection features.

For business leaders, the reporting challenge is equally concrete. Recovery should describe what information remains trustworthy and which processes can resume. Availability figures alone cannot answer whether a company can reconcile its obligations or serve customers accurately.

Treat the Yandex Cloud data loss disclosure as a reason to ask for evidence about the records your organization depends on. Identify the authoritative copies, the people responsible for them, and the last successful restoration. Read future Yandex updates against the same standard: whether they demonstrate usable recovery for customers, including existing workloads. Keep the distinction between infrastructure returning and business information surviving visible in internal decisions. When a provider reports progress, ask what that progress means for your application and its records. Which evidence would let your team say the business has recovered, rather than simply that its servers are running again?

Give every agent the context to do better work

Connect your agents to the knowledge, decisions, and history already organized in remio.

remio currently supports Windows 10+ (x64) and Macs with Apple silicon.

Your AI Partner at Work
Get more done with remio

Plan. Create. Deliver.
All in one place.

bottom of page