Working Paper №1

Global AI Governance: A Structural Diagnosis and Re-Grounding

Harmondeg Institute for Philosophy & Practice

Series: AI Governance Structural Analysis

Date: 2026

Abstract

Across jurisdictions, AI governance has developed through diverse institutional, regulatory, and technical approaches. From classification-based regulation to accountability frameworks and execution-driven oversight, these models reflect significant variation in how governance is conceived and applied.

This paper argues that these differences do not resolve the underlying structural limitation.

AI governance remains positioned at the level of outcomes, acting on systems after they are already in motion. In contrast, AI systems generate behavior dynamically across a continuous chain from design to deployment and ongoing interaction.

This creates a persistent misalignment: governance responds to what appears, while system behavior is formed before it becomes visible.

By synthesizing analyses across the European Union, the United Kingdom, Canada, and Asian models, this paper identifies a common structural condition underlying current governance approaches and outlines the need to reposition governance at the level of system formation.

1. Introduction

AI governance is often framed as a problem of design.

Different jurisdictions adopt different models. Regulatory frameworks evolve. Standards are refined. Accountability mechanisms are strengthened.

These efforts suggest that governance progresses through iteration.

Yet a deeper question remains:

whether governance is positioned at the correct level.

This paper argues that it is not.

AI systems do not produce risk as a fixed attribute. They generate it dynamically across design decisions, model development, deployment environments, and continuous interaction.

When governance is applied after this process, it cannot align with how system behavior is formed.

2. The Structural Position of Governance

Across existing models, governance is anchored at the level of observable outcomes.

Risk is identified once systems are operational. Responsibility is assigned based on measurable effects. Intervention is triggered in response to detected impact.

This creates a consistent orientation:

governance acts on what appears.

System behavior, however, is not produced at the level of appearance.

It forms earlier, through processes that shape how systems function before outcomes emerges.

This establishes the structural condition:

governance operates at the point of visibility, while system dynamics are generated upstream.

3. Structural Expressions Across Systems

This structural condition manifests differently across governance models.

3.1 Classification-Based Systems

In classification-oriented frameworks, risk is categorized into predefined levels.

Governance is applied according to these classifications, with regulatory requirements increasing alongside assessed risk.

This approach provides clarity and consistency.

However, classification assumes that risk can be identified as a stable property of a system.

In evolving AI systems, risk is not static.

It forms dynamically across stages of system development and interaction.

Classification therefore operates on representations of risk that have already emerged, rather than on the processes through which risk is generated.

3.2 Accountability-Based Systems

In accountability-centered models, governance is structured around responsibility assignment.

Actors are identified. Obligations are defined. Compliance is enforced.

This approach strengthens traceability and oversight.

However, responsibility operates at the level of outcomes.

It depends on the ability to attribute effects after they occur.

System behavior, by contrast, emerges through distributed processes that unfold before outcomes become visible.

Responsibility can address what appears but cannot align with how those outcomes are formed.

3.3 Execution-Oriented Systems

In execution-driven models, governance is embedded within system operation.

Enforcement mechanisms ensure that systems behave according to defined constraints. Verification processes confirm compliance in real time.

This approach strengthens control.

However, execution operates on conditions that are already defined.

It enforces alignment with specified parameters but does not determine where those parameters originate.

Governance therefore remains positioned after system formation, ensuring adherence to structures it did not shape.

3.4 Participatory and Adaptive Systems

In open and participatory models, governance evolves through engagement and feedback.

Stakeholders contribute to policy development. Systems are adjusted iteratively in response to observed outcomes.

This approach increases responsiveness and inclusivity.

However, participation operates within the same structural positioning.

Governance engages with systems as they unfold, rather than at the stage where system behavior is initially formed.

Adaptation improves response but does not shift where governance begins.

4. A Shared Structural Limitation

Despite their differences, these models converge on the same structural condition.

Governance is applied after system formation.

This produces a persistent misalignment:

regulation reacts to outcomes, while system behavior is generated upstream.

Variations in institutional design, execution capacity, or participatory openness do not alter this condition. They produce different modes of response, but do not change the structural position from which governance operates.

As a result, governance remains reactive.

It can manage effects but cannot align with the processes that generate them.

5. Structural Re-Grounding

Addressing this limitation requires a shift in the position of governance.

Governance must move from the level of outcomes to the level of formation.

5.1 Governance as a Condition of Formation

Governance cannot remain an external layer applied to systems.

It must become part of the conditions under which systems are formed.

This includes engagement with design, model architecture, and interaction structures.

5.2 Alignment Before Variation

Structural alignment must precede institutional variation.

Differences in regulatory models, execution mechanisms, or participatory processes become effective only when governance is positioned at the stage where system behavior is generated.

5.3 From Response to Alignment

Governance must transition from reacting to outcomes to aligning with system dynamics.

This requires redefining governance as a continuous process across system formation, rather than episodic intervention aftereffects appear.

6. Conclusion

The challenge facing AI governance is not the absence of models.

It is the position from which those models operate.

Across jurisdictions, governance continues to act on systems after they are already in motion.

Without structural repositioning, governance will remain reactive to outcomes it did not shape.

The question is not how governance is implemented.

It is where governance begins.

Related Works (Series)

WP1 — Global AI Governance: A Structural Diagnosis and Re-Grounding

AP1 — EU AI Governance Is Misaligned

https://doi.org/10.5281/zenodo.19297123

AP2 — Re-Grounding AI Governance in the United Kingdom

https://doi.org/10.5281/zenodo.19389558

AP3 — AI Governance in Canada Is Misaligned

https://doi.org/10.5281/zenodo.19391836

AP4 — AI Governance in Asia

https://doi.org/10.5281/zenodo.19421827