Strikes in the Gulf: Lost data serves as a reminder that off-site backups are the client’s responsibility

AWS Declines to Restore Data from Bahrain Data Center Amid Drone Strikes

AWS has announced it will not restore data for certain customers stored in its Bahrain data centers, which were struck by drones. Since March, the provider had advised customers to duplicate their data in another region, emphasizing that the responsibility for backups lies with the client. Nevertheless, it remains unclear why some data stayed in Bahrain without an off-site copy, despite ongoing alerts from the provider and a public threat issued a month prior.

The announcement came via the public health dashboard for AWS services, confirming that it cannot recover resources and data solely hosted in its Bahrain region, ME-SOUTH-1, and the mec1-az2 availability zone in its UAE region, ME-CENTRAL-1. According to AWS, the impact of the attacks on Bahrain affected multiple availability zones and exceeded the capabilities of its regional and multi-zone services. The provider noted that most clients had already restored operations in other regions using backups.

Between the initial drone strikes on March 1 and the assessment on September 15, AWS consistently urged these clients to conduct data backups. However, the company has not disclosed the number of affected clients or the volume of lost data, and the conclusion pertains only to a fraction of the overall issue. In Bahrain, where the second availability zone was compromised in April, some data remained without an off-site copy, despite a month of warnings and a public alert, with neither AWS nor the impacted clients providing clarification.

Six Months of Alerts Preceded Data Loss

On March 1, 2026, AWS reported that objects struck one of its data centers in the UAE, sparking sparks and a fire that prompted firefighters to cut power to the site. The next day, a second availability zone in the same region fell offline, making it impossible to launch new instances. By March 3, AWS announced that both Gulf regions experienced drone strikes, directly impacting two facilities in the UAE and a third strike near a site in Bahrain. The damage included structural and electrical issues, worsened by water from firefighting systems.

Right after the March 3 communication, AWS advised customers in the Middle East to back up their data and consider migrating workloads to other regions, such as the United States, Europe, or Asia-Pacific. In late March, various Iranian authorities made multiple public statements against U.S. tech companies in the Gulf. By April, it became clear that a second zone in Bahrain had also been lost. On April 30, AWS declared the Bahrain region unavailable, suspended billing, and instructed customers to restore resources from remote backups. Clients who ultimately lost everything on September 15 had received this directive nearly six months prior.

The Contract Places Backup Responsibility on Clients

Under Article 2.3 of the AWS Customer Agreement, updated on August 14, 2026, clients are accountable for configuring services and implementing appropriate measures to secure, protect, and back up their accounts and content. This can include encryption and regular archiving. Consequently, AWS was not in a position to act on behalf of its clients. The shared responsibility model applied to resilience differentiates between cloud resilience, which is the provider’s responsibility, and resilience in the cloud, which falls to the client. For managed services like S3 or DynamoDB, AWS’s documentation explicitly places backup, version management, and data replication within the client’s domain.

The S3 storage service is designed to withstand the total loss of an availability zone, as emphasized by AWS on March 2, following the first strike. The provider’s commitment pertains to this architecture. The loss of a second zone exceeded that framework. The contract also stipulates, under Article 11.3, a force majeure clause covering acts of war and caps AWS’s liability at the amounts paid in the previous twelve months. Thus, clients without an off-site copy had neither data to restore nor recourse to recover their losses adequately.

High Availability Is Not a Disaster Recovery Plan

Clients who only distributed their applications between the Bahrain zones lost access to their data in April when the second zone failed, followed by the actual data loss on September 15. The Gulf situation highlights a common confusion in cloud architectures. Distributing an application across multiple availability zones protects against the failure of a data center. However, in the event of a total regional loss, only an off-site copy—either in another region or with a different provider—can facilitate recovery. The warning period was insufficient for all clients, and this situation remains unexplained.

The costs associated with off-site backups, including storage, data transfer, and restoration testing, can often lead enterprises to make budgetary trade-offs. AWS’s shared responsibility model offers IT managers clarity regarding responsibilities and allows clients to choose the location of their backups. However, it also places the financial burden solely on the enterprise, particularly for a risk deemed improbable. Additionally, stringent data localization requirements common in the Gulf may compel companies to store everything in one region. AWS’s migration recommendation included caveats regarding latency and data residency requirements.

AWS will re-evaluate the situation in Bahrain in early 2027 and is replacing the damaged infrastructure in the UAE. For European businesses, the key takeaway lies in their own contracts. Mapping out the regions where their backups reside has become a critical aspect of risk management, akin to vendor mapping.

Scroll to Top