The classic trap
Article 83 targets formal non-compliance: not a substantive defect (a biased algorithm, a security flaw), but the absence of mandatory administrative evidence. This is the most insidious trap, because a high-risk AI system can be perfectly engineered on a technical level and still be illegal because the CE marking is missing, the EU declaration of conformity was never drawn up, or the registration in the EU database (Article 71) was not carried out. The competent market surveillance authority (in Luxembourg, not yet formally designated to date, with the EU AI Office supervising general-purpose models) grants a compliance period, then, if the non-compliance persists, restricts, prohibits, recalls or withdraws the system from the market. A purely declarative defect is enough to trigger the withdrawal.
The 7 formal failures that get your system withdrawn from the market
Article 83(1) precisely lists what the authority checks first during an inspection, even before analysing model performance:
- The CE marking affixed in violation of Article 48 (wrong format, placement, or affixed without legal basis).
- The CE marking simply absent.
- The EU declaration of conformity (Article 47) never drawn up.
- The EU declaration of conformity drawn up incorrectly or incompletely.
- The lack of registration in the EU database under Article 71.
- No authorised representative appointed when the provider is established outside the EU.
- Technical documentation not available or not presentable on request.
Each of these points is binary: either the evidence exists and is up to date, or it is missing. There is no arguable grey zone, which makes the exposure particularly brutal for providers who manage their evidence in scattered shared folders.
How Luxgap automates this risk
Our Luxgap CE Conformity Vault makes withdrawal for formal non-compliance impossible by turning the Article 83 checklist into an evidence vault always ready for an inspection. The tool continuously monitors the existence, validity and consistency of every mandatory artefact (CE marking, EU declaration of conformity, EU database registration, mandate, technical documentation) by connecting to your GitLab, SharePoint M365, Confluence repositories and your PLM, without asking your compliance officer to maintain a manual spreadsheet.
- Automatically detects each high-risk AI system as soon as a new project appears in your GitLab, Azure DevOps repositories or your MLflow model catalogue.
- Verifies in real time the presence of a signed, dated Article 47 EU declaration of conformity matching the regulatory template, and flags missing or inconsistent fields.
- Checks the effective registration in the EU database under Article 71 and flags any system placed on the market without a corresponding entry.
- Scans the completeness of the technical documentation required by Annex IV and computes a presentability score before inspection.
- Instantly alerts via Teams or Slack whenever a required authorised representative for a non-EU provider has not been appointed.
- Produces a timestamped, sealed PDF report, enforceable before the market surveillance authority, demonstrating that every formal requirement of Article 83 is met as of the inspection date.
Available as a complement to a Luxgap DPO or CISO mandate or as a dedicated SaaS module depending on your scope. Request a tailored quote and our teams will prepare a demonstration on your real AI systems, with a free blind audit within 48h to measure your formal exposure before any commitment.
