The AI Act deadline moved. The documentation job didn’t.
The sensible first question is not “Are we compliant?” It is “What evidence do we have, what is missing, and who owns the gaps?”
The European Union has adjusted parts of the AI Act timetable. That matters for planning. It does not make the underlying documentation work disappear.
Article 11 requires technical documentation for high-risk AI systems to be prepared before the system is placed on the market or put into service and kept current. Annex IV describes the information that record should contain. The list reaches well beyond a model card: intended purpose, development methods, data, performance, monitoring and control, risk management, lifecycle changes, standards, the declaration of conformity, and post-market monitoring.
A checklist is not the same as evidence
Teams often begin with the right headings and then discover that the supporting material is scattered across engineering tickets, risk registers, test exports, vendor documents, and someone’s memory. A heading marked “complete” does not tell a reviewer which version was tested, why a metric was chosen, what changed later, or who accepted the remaining risk.
The useful unit of work is a traceable record. For each documentation area, a team should be able to point to the source, name the owner, identify the system version, and explain what a reviewer can and cannot conclude from it.
Start with an evidence map
I built a small browser-private mapper around the nine documentation areas in Annex IV. It asks you to mark each area as available, in progress, missing, or not applicable. The result is a printable preparation list. Nothing is submitted to VeriTrooper.
The mapper does not decide whether your system is high-risk. It does not provide legal advice, perform a conformity assessment, or certify compliance. Those decisions belong with qualified legal and assurance professionals. Its job is narrower: make the evidence gaps visible before a formal review turns them into expensive surprises.
Where testing fits
For a document-grounded AI workflow, repeatable answer testing can support parts of the record: the protocol, item-level results, chosen metrics, limitations, exclusions, configuration, and change history. It cannot fill every legal or organizational requirement. It should preserve the reviewer’s judgment, not quietly replace it.
That is the point of VeriTrooper’s 10-business-day AI Evidence Readiness Sprint. Bring us one real AI workflow and the source material it should follow. We will test it, show what holds up or falls short, and give your team or outside reviewer a clear record to use.