AIKNOCK Open specification · v0.4 Ref. AIKNOCK / WD-0004 · 11 May 2026

AIKNOCK · Rationale of the protocol

Why an execution-control system for AI.

AIKNOCK is created with a precise objective: to bring AI execution control down to the infrastructure layer, where decisions become technical constraints applied by the system and not mere statements of intent.

§ 01 · The idea in one sentence

The idea in one sentence.

Every invocation of AI traverses the system. Its execution behaviour is constrained by construction, not by intention.

§ 02 · The problem

The problem everyone sees, but few address.

AI invocations today originate from heterogeneous applications, written by distinct teams, directed at different models and providers. Such invocations transit without a technical control point interposed at system level.

Invocation control, however, almost always remains:

This approach presupposes the cooperation of applications. But when a technology becomes critical, pervasive and decisional, trust in the invoking application is not a technical strategy. The problem addressed is technical and concerns the behaviour of the computing system during the execution of artificial intelligence operations.

§ 03 · Historical lesson

The lesson from the history of computing.

The internet did not become reliable by asking applications to behave. It became reliable when TCP/IP moved control beneath the applications, into the network protocol.

Network · TCP/IP

Moved control beneath the applications, into the network protocol.

AI · AIKNOCK

Move control beneath the applications, into the execution-control protocol.

AIKNOCK applies the same principle to artificial intelligence:

§ 04 · What it is

What AIKNOCK is.

AIKNOCK is an ex-ante execution-control system, designed to be integrated at the level of the operating system, the execution runtime and the base platform.

Every invocation of AI must traverse a mandatory, non-bypassable control point, before the model is executed. The system evaluates and constrains AI usage before execution.

§ 05 · Key principles

The key principles.

  1. Mandatory interposition.

    AI is not a free resource. It is a capability mediated by the system.

  2. Decision before action.

    Context and intent are evaluated before AI is invoked, not afterwards.

  3. Technical enforcement.

    The system can allow, limit, degrade or block the use of AI at runtime, independently of the application.

  4. Audit by construction.

    Evidence is not an optional log: it is an inevitable effect of how the system operates.

  5. Human-in-the-Loop as a constraint.

    Where needed, human intervention is not a procedure but a technical requirement enforced by the system.

§ 06 · Boundaries

What AIKNOCK is not.

For clarity, AIKNOCK is not:

AIKNOCK is infrastructure, not application.

§ 07 · De-facto standard

Why a de-facto standard.

AIKNOCK is not created to be "chosen". It is created to become inevitable in contexts where AI is critical.

Like TCP/IP, like identity mechanisms, like secure boot:

  1. Stage 01

    first it works,

  2. Stage 02

    then it is adopted,

  3. Stage 03

    finally it becomes a standard.

§ 08 · Vision

A concrete vision.

AI execution control cannot remain an act of trust.

It must become a technical property of information systems.

AIKNOCK exists to make this vision realisable.