The classic trap
Recital 7 sets the interpretive lens for the ENTIRE AI Act: 'high-risk' rules are not a mere technical checklist, they must be read in light of the Charter of Fundamental Rights, the AI HLEG ethics guidelines and the European Declaration on Digital Rights. The classic trap is to treat AI Act compliance as a documentary exercise (FRIA + Annex IV technical file) while ignoring the fundamental rights dimension. The EU AI Office and, for the personal data leg, the CNPD expect a reasoned demonstration that your system does not impair health, safety, non-discrimination, dignity or privacy, not a ticked box.
The 'public interest' test: the 4 axes of argumentation expected
When a regulator reads your AI compliance file, they look for your reasoning along four axes drawn directly from recital 7. Each must be documented with concrete evidence, not generic prose:
- Health: direct or indirect impact on physical or psychological integrity (medical triage, health insurance scoring, sensitive content moderation).
- Safety: robustness against errors, adversarial attacks, model drift, and the consequences of a failure.
- EU Charter fundamental rights: dignity (art. 1), non-discrimination (art. 21), privacy (art. 7-8), freedom of expression (art. 11), effective remedy (art. 47).
- Alignment with the AI HLEG: human agency and oversight, technical robustness, privacy, transparency, diversity and fairness, societal well-being, accountability.
A high-risk system whose technical file does not explicitly address any of these axes is a file that does not stand. The Article 27 FRIA then becomes the pivot tool that materialises this argumentation.
How Luxgap automates this risk
Our Luxgap Fundamental Rights Compass turns the interpretive requirement of recital 7 into an opposable argumentation matrix. Rather than asking you to fill a blank FRIA template, the tool queries your connected systems (AI register in Odoo or ServiceNow, models deployed on Azure ML / AWS SageMaker / Vertex AI, referenced training datasets) and generates a structured argumentation file on the 4 axes of recital 7, cross-referenced with the 7 AI HLEG principles.
- Automatically maps every AI system deployed across your IT estate via Azure ML, SageMaker, Vertex AI, Hugging Face Hub and internal registry connectors.
- Evaluates exposure against the 7 AI HLEG principles and produces an alignment score per system, with ready-to-insert justification text for the Annex IV technical documentation.
- Detects non-discrimination blind spots by analysing training datasets (gender, age, origin, postcode imbalances) without extracting the data outside the GDPR perimeter.
- Generates an Article 27 FRIA pre-filled and explicitly tied to the relevant Charter articles and ethics guidelines.
- Produces a timestamped PDF report, opposable to the EU AI Office and the CNPD, demonstrating reasoned consideration of the public interests of recital 7.
Available as a complement to a Luxgap DPO or CISO mandate or as a standalone SaaS module depending on your perimeter. Request a tailored quote and our teams will prepare a demonstration on your actual models, with a free 48h blank audit to size your exposure before any engagement.
