Assessment Report "Not a High-Risk AI System" – nori
This translation is NOT legally binding and a working translation only. Legally binding and relevant, particularly in case of any discrepancies, is solely the German text.
Subject: nori (AI-supported compliance analysis engine) Legal basis: Art. 6(3) and (4) AI Act (Regulation (EU) 2024/1689), Recital 65 Version date: 17 August 2026 · Status: Approved
Purpose: documentation of the decision and the reasoning why nori is not to be classified as a high-risk AI system (Art. 6(4) AI Act).
1. Result
nori is not a high-risk AI system within the meaning of the AI Act. The reasoning proceeds in two steps:
- Primarily: nori falls under neither Annex I nor Annex III of the AI Act.
- In the alternative: even assuming a link to Annex III, the exemption conditions of Art. 6(3) apply (narrow procedural/preparatory task, human oversight, no profiling).
nori is also not a prohibited practice under Art. 5.
2. System description (relevant to the decision)
- Function: static, AI-supported analysis of source code against regulatory/security frameworks; output of textual findings/reports.
- Role in the decision-making process: nori delivers suggestions/indications. The regulatory assessment (pass/fail, approval) is made by a human auditor.
- Affected parties: software artefacts and development/audit teams. No automated decisions about natural persons.
- No profiling of natural persons; no processing of personal data as the subject of assessment (except for data that may incidentally be contained in the source code – not a purpose of the assessment).
3. Assessment under Art. 5 (prohibited practices)
None of the prohibited practices (including subliminal manipulation, social scoring, remote biometric identification, emotion recognition in the workplace, exploitation of vulnerability) applies to nori. → Not applicable.
4. Assessment under Annex I (safety component of regulated products)
nori is not a product and not a safety component of a product falling under the harmonisation legislation (Annex I). It is a standalone development/audit tool and is not embedded in a regulated product. → Not applicable.
Note: the fact that nori is used to audit health software (e.g. DiGA/DiPA), among other things, does not make nori itself a medical device or a safety component thereof. nori is part of the development/assessment process, not of the product being assessed.
5. Assessment under Annex III (high-risk areas of application)
| Annex III area | Applies to nori? | Reasoning |
|---|---|---|
| 1. Biometrics | No | No biometric identification/categorisation/emotion recognition. |
| 2. Critical infrastructure | No | nori is not a safety component in the operation/management of critical infrastructure. |
| 3. Education and vocational training | No | No assessment of learners, no access to or allocation of education. |
| 4. Employment/workforce management | No | No recruitment, selection or performance evaluation of individuals. |
| 5. Access to essential services | No | No creditworthiness/eligibility assessment; no decisions on access to public or private services for individuals. |
| 6. Law enforcement | No | No use by law enforcement authorities, no risk or evidence assessment. |
| 7. Migration/asylum/border control | No | No connection. |
| 8. Justice/democratic processes | No | No support for judicial authorities, no influence on elections. |
→ No Annex III area is applicable.
6. In the alternative: assessment under Art. 6(3)
Even if (contrary to section 5) a link to Annex III were assumed, nori would not be high-risk under Art. 6(3), because it fulfils one of the conditions set out there:
- (a) narrow procedural task – nori assesses code against clearly delimited controls; and
- (d) preparatory task to an assessment carried out by humans – findings prepare the human compliance review, but do not replace it.
The exclusion clause of Art. 6(3) (profiling of natural persons) does not apply, as nori does not carry out profiling. The exemption would therefore be applicable.
7. Risk considerations and mitigation measures
| Measure | Implementation in nori |
|---|---|
| Human oversight | Findings = suggestions; the auditor decides. Feedback/false-positive annotation in the dashboard. |
| Traceability | run.json per run (model, git commit/branch, timestamp, score, tokens/cost); findings with file/line references. |
| Quality control | nori-regression: precision/recall/verdict accuracy against committed ground truth + baseline, with human adjudication. |
| Determinism/stability | Template-based prompts; deterministic integrity tests in CI. |
8. Conclusion, responsibilities, re-assessment
Conclusion: nori is not to be classified as a high-risk AI system (Art. 6(3)/(4)). This assessment is to be registered before placing on the market/putting into service and made available on request.
Responsible: nori product owner Dr. Matthias Berger
Events triggering a re-assessment:
- substantial change in functionality (e.g. autonomous approval decisions without human oversight),
- use in an Annex III context,
- processing/profiling of natural persons,
- changes to the AI Act or to relevant guidelines.