Germany's Federal Office for Information Security (BSI) published a Community Draft of its AI Audit and Assurance Assessment Architecture (A5) on July 6, 2026 — a modular framework of criteria and methodology for assessing the trustworthiness of AI systems across the value chain, from developers to operators to procurement officers. Public comment is open through August 31, 2026, submitted via email to aisecurity@bsi.bund.de (BSI, A5 node).
What A5 Actually Does
A5 is not a new law. It is an audit architecture: a horizontal "base module" of technology-agnostic criteria, extendable with sector- or application-specific modules, published in both document form and machine-readable OSCAL format so auditors can slot it into existing compliance tooling. Its methodology borrows directly from BSI's C5 Cloud Computing Compliance Criteria Catalogue, using the internationally recognized ISAE 3000 assurance standard rather than inventing a bespoke German procedure. BSI is explicit that the goal is to help organizations document technical trustworthiness against "current and upcoming regulatory conditions, such as the EU AI Act or Cyber Resilience Act" (BSI, A5 node).
This is not BSI's first attempt at this. It already runs the AI Cloud Service Compliance Criteria Catalogue (AIC4), a voluntary extension of C5 covering explainability, robustness, and bias in machine-learning cloud services, and it published a criteria catalogue for AI systems in the financial sector earlier in 2026 (BSI, AIC4 node). A5 generalizes that pattern into a single modular architecture meant to cover the whole AI value chain, not just cloud-delivered models or one regulated sector.
Filling a Vacuum Brussels Just Widened
The timing is not incidental. On May 6, 2026, the Council and European Parliament reached political agreement on a "Digital Omnibus" that pushes back the AI Act's high-risk system obligations: stand-alone Annex III systems (recruitment, credit scoring, law enforcement, education tools) now face a compliance deadline of December 2, 2027, and AI embedded in regulated products under Annex I moves to August 2, 2028 — both roughly a year and a half later than the original schedule (Gibson Dunn). Article 43's mandatory conformity assessment regime, in other words, will not bind most high-risk AI providers for another eighteen months.
As a German industry analysis of BSI's parallel financial-sector catalogue put it, "die EU-KI-Verordnung steht, aber vielen fehlt der Hebel für die Praxis" — the regulation exists, but many lack the practical lever to apply it (Bitkom Consult). That is the gap BSI is stepping into: giving industry a concrete, testable standard to build toward voluntarily, well before Brussels' own mandatory clock starts running.
The Case For It
The strongest argument for A5 is that certification infrastructure takes years to mature, and starting now — while adoption is voluntary and low-stakes — is far better than scrambling to build audit capacity in the eleven months before a mandatory deadline. Enterprises procuring AI systems, insurers pricing AI liability, and public bodies buying AI tools all currently lack a shared, technically rigorous way to compare vendor trustworthiness claims. A modular, OSCAL-native standard built on the already-established C5/AIC4 lineage gives the market a common vocabulary and reusable audit evidence, rather than each customer inventing its own due-diligence checklist. This also mirrors how Germany handled NIS2: after missing the EU's October 2024 transposition deadline by more than a year, its NIS-2-Umsetzungsgesetz finally took effect on December 5, 2025, immediately binding roughly 29,500 companies with no grace period (Global Policy Watch). A voluntary AI trust framework, refined through an open comment process before it hardens into anything binding, is a more measured approach than repeating that pattern of legislate-late-then-impose-abruptly.
Why Germany Should Keep the Word "Voluntary" Load-Bearing
The risk is not A5's substance — a technically serious, OSCAL-based, ISAE 3000-grounded methodology is sound engineering. The risk is drift. Germany has a track record of national bodies producing frameworks that start voluntary and end up functioning as de facto market requirements once large public buyers or insurers start demanding them as a condition of doing business — effectively a certification mandate without ever passing through legislative scrutiny or a proportionality test. For a Mittelstand-heavy AI sector where compliance capacity is scarcer than at hyperscalers, an audit architecture that quietly becomes table stakes for public procurement would function as a barrier to entry precisely while the EU's own conformity assessment regime is on hold.
BSI's own framing — that A5 helps meet "upcoming" regulatory conditions rather than current binding ones — is the right posture, and the extended comment window through August 31 is a genuine opportunity to keep it there. The sensible outcome is a widely adopted voluntary standard that reduces duplicative one-off audits and gives the AI Act's delayed Article 43 process something mature to draw on when it finally activates in December 2027 — not a shadow certification regime that forecloses the market experimentation the delay was supposed to allow.