The classic trap
Recital 171 illuminates Article 86 of the AI Act: it establishes a genuine right to explanation for persons affected by an automated decision based on a high-risk AI system. The practical trap is twofold: first, many deployers believe that a generic GDPR Article 22 mention is sufficient, whereas the AI Act requires a clear and meaningful explanation specific to the model's output; second, the EU AI Office and the CNPD (on the personal data side) expect an actionable explanation, meaning one that concretely enables the person to exercise their rights, and not a copy-paste of the technical model card.
What a recital 171-compliant explanation must contain
- Identification of the high-risk AI system involved and its exact role in the decision (determinative, contributive, secondary).
- The main input variables that drove the output for this specific person (not the model's global explainability).
- The general logic of the automated processing and any thresholds or business rules applied by the deployer downstream of the AI output.
- Available remedies: contestation, human review, complaint to the competent authority.
- Traceability of the exact model version used at the time of the decision (model card, hash, deployment date).
- Documented exclusion cases (public security, regulated anti-fraud) with the precise legal basis cited.
The practical test: clear AND meaningful
A purely technical explanation (SHAP values, feature weights) does not satisfy recital 171. Conversely, a vague formula (your file does not meet our criteria) is insufficient too. The expected standard is that of a reasonable person who, after reading, understands why they received this decision and what they can do to contest it.
How Luxgap automates this risk
Our Luxgap Decision Explainer transforms each output of your high-risk AI system into an individualised explanation, readable by the affected person and defensible before the EU AI Office or the CNPD. The tool sits between your AI engine (Azure ML, AWS SageMaker, Vertex AI, Dataiku, in-house models) and your client notification channel (Salesforce Service Cloud, Zendesk, Odoo portal, postal mail via an editing solution), and generates an individual timestamped explanation letter for each automated decision on the fly.
- Captures the model's raw output and the exact deployed version at the moment of the decision, with a cryptographically signed hash for downstream evidence.
- Translates technical variables (features, scores, thresholds) into business-readable language understandable by the person, in their language (FR, EN, DE, LU, PT).
- Detects whether the decision falls under a legal exclusion (CSSF anti-money-laundering, public security) and adapts the explanation content accordingly, citing the legal basis.
- Generates a timestamped PDF with available remedies clearly listed (internal contestation, human review, EU AI Office complaint, CNPD recourse for the data side).
- Alerts the DPO and the AI officer in real time via Teams or Slack when an explanation request arrives, with the legal response deadline pre-computed.
- Maintains an audit-grade log of every explanation delivered, exportable on demand by the AI market surveillance authority.
Available as a complement to a Luxgap DPO or CISO mandate or as a standalone SaaS module depending on your scope. Request a tailored quote and our teams will prepare a demonstration on one of your real high-risk models, with a free 48h baseline audit to measure the current quality of your explanations before any engagement.
