The EU's Digital Omnibus, agreed in May and formally adopted this summer, pushed the AI Act's high-risk obligations sixteen months into the future: standalone Annex III systems now face a deadline of 2 December 2027, and AI embedded in regulated products under Annex I moves to 2 August 2028. The Commission cited a straightforward reason — CEN-CENELEC could not deliver harmonized standards in time, and without them, conformity assessments are unenforceable paper exercises. What the Omnibus did not do was tell member states how to fill the gap.
Germany is not waiting. On 6 July 2026, the Federal Office for Information Security (BSI) published a community draft of its AI Audit and Assurance Assessment Architecture (A5): a modular, extensible framework of criteria and methodology for assessing AI trustworthiness across the entire value chain — developers, operators, oversight bodies, and procurement officers. Public comment closed on 31 August, which means the BSI is now reviewing feedback with a view to finalizing what will likely become the most concrete national AI compliance architecture in Europe.
A5 is technically serious. Its criteria align with BSI's established "State-of-the-Art Library" conventions, are published in both document and machine-readable OSCAL formats, and its audit methodology borrows directly from the internationally recognized ISAE 3000 assurance standard. The framework is explicitly designed to slot into existing compliance toolchains and to address "current and upcoming regulatory conditions, such as the EU AI Act or Cyber Resilience Act." This is not a hobbyist checklist; it is an attempt to build certification infrastructure while Brussels' own mandatory clock is paused.
The case for it is pragmatic. Certification ecosystems take years to mature. Starting now — while adoption is voluntary and stakes are low — beats scrambling to build audit capacity in the eleven months before a mandatory deadline. German enterprises, insurers, and public buyers currently lack a shared, technically rigorous way to evaluate vendor trustworthiness claims. A5 gives them a common vocabulary and reusable audit evidence.
But the risk is drift. Germany has a track record of national bodies producing frameworks that begin as voluntary guidance and end up as de facto market requirements once large public buyers or insurers demand them as a condition of doing business — effectively a certification mandate that never passed 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 when the EU's own conformity assessment regime is on hold.
DigitalEurope, the industry association representing major technology firms, has been explicit about this danger: one of its stated policy priorities is to "support the Commission's goal to avoid the development of national initiatives that would further fragment the European regulatory environment." The tension is real. The Commission retains its Article 41 power to adopt common specifications, but it has not used it. Meanwhile, a technically mature German standard is consolidating just as the delay was supposed to create breathing room.
Whether A5 becomes a widely adopted voluntary scaffold that the eventual harmonized standards can draw upon — or a shadow regime that fragments the single market before the Commission acts — depends less on the document's technical quality, which is sound, than on whether Germany keeps the word "voluntary" load-bearing.
Sources
– BSI, "AI Audit and Assurance Assessment Architecture (A5)"
– People of Internet, "Germany's BSI Moves to Fill the AI Act's Assurance Gap With a Voluntary Audit Standard"
– Regulation AI, "EU AI Act Omnibus 2026: New Compliance Deadlines Explained"
– Gibson Dunn, "EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes"
– DIGITALEUROPE, "Artificial Intelligence & Data" policy page