Norms and AI execution control
A technical reading of the relations among AI Act, NIS2, ISO/IEC 27001, ISO/IEC 42001 and NIST AI RMF, and of the role a system-level execution-control system may take with respect to the requirements they pose.
§ 01 · Premise
The AI Act, NIS2 and the ISO and NIST standards operate on different layers: legal, organisational, risk-management. Although heterogeneous, they converge on a restricted set of key requirements.
Read by principles rather than by articles, these requirements reduce to four recurring properties: preventive control of AI invocation, documented and verifiable traceability of decisions, technical auditability and accountability attributable to identifiable subjects.
The legal layer defines obligations; the organisational layer defines processes; the risk-management layer defines methods. Missing from this stratification is an explicit definition of the technical mechanisms through which such requirements can be made observable in the systems that actually execute AI.
§ 02 · Reference frameworks
The table below summarises, in conceptual and non-legalistic form, the principles that each normative or framework reference makes explicit. It does not constitute a conformity analysis nor a binding interpretation of the texts.
| Framework | Layer | Relevant principles |
|---|---|---|
| AI ActReg. (EU) 2024/1689 | Legal | Ex-ante control of the use of AI systems; human oversight on decisions of significant impact; auditability of systems across the life cycle. |
| NIS2Dir. (EU) 2022/2555 | Systemic security | Risk management of critical capabilities, continuity and security of essential services, tracking and notification of incidents. |
| ISO/IEC 27001Information security management systems | Organisational | Control of access to critical resources, separation of privileges, recording and review of activities. |
| ISO/IEC 42001Artificial intelligence management systems | Organisational | Management system for AI systems: traceability of decisions, roles and responsibilities, periodic review. |
| NIST AI RMF 1.0AI Risk Management Framework | Risk management | Functions Govern, Map, Measure, Manage: direction, identification of the use context, measurement and treatment of risk. |
§ 03 · Missing denominator
The requirements listed in the reference frameworks are expressed in terms of what must be guaranteed. Their operational translation is today entrusted to heterogeneous instruments — textual rules, documentation, application controls, after-the-fact records — rarely brought back to the same architectural layer.
In particular, a technical control point is missing that jointly presents the following four properties.
The absence of such a control point is not a gap in the norms: it is a gap in the architecture of the systems. It is, in other words, an architectural problem.
§ 04 · Position
AIKNOCK explores a possible ex-ante technical control point, at the system layer, consistent with the execution-control, auditability and risk-management requirements referenced by norms and standards, without replacing or interpreting them.
Norms and standards define obligations. AIKNOCK operates on the layer of technical mechanisms. The system operates independently of external regulatory or standard frameworks and maintains a consistent technical behaviour at execution level. The two layers are complementary: what the law and the standards prescribe as system properties, a mechanism such as the one described can contribute to making observable and verifiable.
§ 05 · Exclusions
To avoid confusion between distinct layers, it is appropriate to state explicitly the scope that AIKNOCK does not occupy.
Observance of the norms remains a responsibility of operators, implementers and competent authorities. AIKNOCK does not shift such responsibility: it describes a distinct technical layer.
§ 06 · Conclusion
The norms define what must be guaranteed.
AIKNOCK sits on the layer of how a part of such guarantees can be made a technical property of the system.
The distance between the two layers is a structural feature, not an opposition: the two levels presuppose each other.