BSI sets cloud evidence framework for AI systems

The Federal Office for Information Security (BSI) has published the A5 Cloud Infrastructure Operations Module for cloud-based AI systems, dated 3 August 2026. [Source: BSI, A5 Cloud Infrastructure Operations Module, PDF, publication details]

The PDF forms part of the AI Audit and Assurance Assessment Architecture. BSI describes it as a vertically oriented module that supplements the A5 Horizontal Trustworthiness Core Module with requirements arising from operating a cloud-based AI system. [Source: BSI, A5 Cloud Infrastructure Operations Module, relevant section]

The module addresses infrastructure and essential system and platform components supplied through cloud services. It provides a basis for demonstrating the trustworthiness of that infrastructure. It does not create a new AI-risk category or directly regulate a foundation model.

The document therefore sits in the cloud-infrastructure and deployed-system layer of the foundation-model versus deployed-system split. It does not classify a general-purpose AI model, establish GPAI obligations, or apply the Article 6 high-risk test to a particular use case. Those questions remain separate from evidence showing that the hosting environment operates reliably and securely.

The accompanying BSI material identifies Section GCS.2.5 as addressing recognised evidence, including a C5 certificate or report covering relevant infrastructure, platform, and system components and risk dimensions. [Source: BSI, A5 Cloud Infrastructure Operations Module, Section GCS.2.5]

For a product deployer, that points to an evidence trail: identifying the cloud components supporting the AI service, logging the relevant assurance documents, mapping their coverage to operational risks, and recording how gaps are handled.

The user-facing consequences sit behind those controls. Availability evidence can support service-continuity decisions, security evidence can inform deployment approvals, and infrastructure reporting can give incident teams a defined record of affected components. The module does not specify a new reporting channel, conformity deadline, or affected-firm count.

The BSI architecture separates the horizontal trustworthiness core from operational modules for particular environments. That structure gives assessors and deployers a way to calibrate model-card disclosure against the infrastructure evidence supporting the deployed service, without treating cloud assurance as content moderation or model governance.

The publication is BSI guidance, not an EU AI Act implementing act or a binding legal requirement identified in the briefing. It also does not resolve the wider standardisation gap involving bodies such as CEN-CENELEC JTC 21 or ISO/IEC JTC 1/SC 42. [Sources: CEN-CENELEC JTC 21, official webpage; ISO/IEC JTC 1/SC 42, official webpage]

Cloud providers, AI system deployers, and assurance assessors would use the module when assembling evidence against the A5 architecture. BSI published it on 3 August 2026 as a 1MB PDF download, and the source does not specify a mandatory legal process attached to its use. [Source: BSI, A5 Cloud Infrastructure Operations Module, official PDF download page]