On the Persistence of Conceptual Errors in AI Governance and the Necessity of Reattaching Responsibility

 ADDENDUM TO :

PRINCIPLE OF RESPONSIBILITY OF NEW INTELLIGENCES II - English translation - Foundations, Analogies, and Transcending Classical Regimes

By

Patrick Houyoux, LL.M.
President — PT SYDECO
Sovereign Cybersecurity & Artificial Intelligence

“ARDENS DEFENDERE”

 

April 2026

 

Author:

Patrick Houyoux, LL.M.

President & Founder — PT SYDECO

Recognized as Best IT Security Business Leader 2026 (Indonesia) — APAC Insider

Specializing in:

• Sovereign Cybersecurity Architectures
• AI Governance & Responsibility
• Legal Structuring of Emerging Technologies

 

Institutional Positioning

PT SYDECO develops sovereign, on-premise cybersecurity and AI systems,
with a focus on:

• Edge-layer prevention
• Legal-grade evidence and traceability
• Independence from foreign cloud infrastructures

 

Doctrinal Series

This document forms part of the series:

“The Principle of Responsibility of New Intelligences”

• Part I — Foundational Principle
• Part II — Foundations, Analogies, and Transcending Classical Regimes
• Part II.1 — Addendum (this document)

THE DOCUMENT

On the Persistence of Conceptual Errors in AI Governance and the Necessity of Reattaching Responsibility

 

Abstract

 

The rapid expansion of artificial intelligence governance frameworks has not been accompanied by a corresponding clarification of responsibility. On the contrary, contemporary discourse increasingly relies on notions such as “distributed accountability,” which, while descriptively accurate at an organizational level, are conceptually incompatible with the legal structure of responsibility.

 

This addendum identifies and corrects the principal errors underlying current approaches. It argues that artificial intelligence does not transform the nature of responsibility but rather obscures its attribution. Consequently, responsibility must be reaffirmed and reattached to the human actors who organize, deploy, and control such systems.

 

I. The Illusion of “Distributed Accountability”

A recurring assertion in contemporary AI governance discourse is that accountability is “distributed” across systems, actors, and processes. This reflects an observable organizational reality: multiple stakeholders intervene at different stages of the lifecycle of an AI system.

However, this descriptive observation cannot be elevated to a normative principle.

From a legal standpoint, responsibility cannot be meaningfully distributed if such distribution results in the impossibility of identifying a responsible subject. Responsibility presupposes attribution. It requires a subject capable of bearing legal consequences.

Thus, what is described as “distributed accountability” is, in reality: a fragmentation of operational roles without corresponding legal attribution.

Where attribution disappears, responsibility does not become collective; it becomes absent.

 

II. The Confusion Between Governance and Responsibility

A second major error lies in the conflation of governance mechanisms with responsibility itself.

AI governance frameworks typically include:

• validation procedures,
• data governance policies,
• monitoring and auditing mechanisms,
• risk management controls.

These elements structure systems. They do not establish responsibility.

Responsibility is not a function of procedural robustness. It is a matter of legal imputation.

Accordingly: A system may be perfectly governed and yet remain legally irresponsible.

Governance organizes action.

Responsibility attributes consequence.

The confusion between the two produces frameworks that are operationally sophisticated but juridically insufficient.

III. The False Autonomy of AI Systems

Contemporary discourse frequently attributes a form of autonomy to AI systems.

This assumption must be rejected in legal analysis.

Artificial (NEW) intelligence systems do not possess:

• legal personality,
• will,
• intention.

They operate through:

• prior human design,
• training and configuration,
• deployment conditions,
• parameterized constraints.

Thus:

AI does not act; it executes an organized structure of human decisions.

The appearance of autonomy is technical, not juridical.

Responsibility cannot be displaced on the basis of such an appearance.

IV. The Fragmentation of the Chain of Responsibility

AI systems involve multiple actors:

• designers,
• developers,
• deployers,
• operators.

Each contributes to system behaviour.

This results in a fragmentation of roles.

However: technical fragmentation cannot justify legal indeterminacy.

The current structural issue is clear: Responsibility is technically fragmented, but legally unassigned.

This is incompatible with the fundamental requirements of legal systems, which demand that:

• actions be attributable,
• consequences be imputable,
• liability be enforceable.

 

V. Reaffirmation of the Principle

The governing principle must therefore be reaffirmed:

“No intelligence, artificial or otherwise, can exist without responsibility.

That responsibility must be attached to the human actors who organize, enable, and control the system”.

 

Responsibility is not created by artificial intelligence.

It is obscured by it.

The central question is therefore: Where must responsibility be attached within the chain of organization?

This requires a systematic reconstruction of responsibility from:

• conception
• design
• deployment
• execution
• outcome

VI. Consequences for Law, Compliance, and Governance

The persistence of conceptual errors produces tangible consequences:

1.     Regulatory insufficiency
Frameworks lacking clear attribution fail to meet legal expectations.

2.     Organizational exposure
Ambiguity in responsibility increases liability risk.

3.     Compliance inadequacy
Process-based compliance without attribution is incomplete.

 Thus, organizations must move beyond process validation and demonstrate: traceability of responsibility

This implies:

• explicit designation of responsible actors
• linkage between decisions and accountable persons
• documentation of authority and control

 

VII. The Risk of Organized Irresponsibility

A critical systemic risk emerges: organized irresponsibility

Defined as: a structural configuration in which decisions are produced, consequences occur, yet no actor can be held fully responsible.

This is not a mere operational flaw.

It is a legal defect in system architecture.

Such systems undermine:

• accountability
• enforceability
• legal coherence

Preventing organized irresponsibility must therefore be a primary objective of AI system design.

VIII. The Necessity of External Responsibility Attribution

In practice, certain sectors have developed mechanisms to address the fragmentation of responsibility inherent in complex systems.

A recurring approach consists in presenting to the user or contracting party a single identifiable entity responsible for the system’s operation and outcomes, notwithstanding the multiplicity of actors involved in its design and functioning.

At first glance, this may appear as a purely commercial or contractual simplification.
In reality, it responds to a deeper legal necessity.

The external unification of responsibility is a condition for its effective existence.

Where multiple actors contribute to a system, internal arrangements may distribute roles, obligations, and liabilities among them. However, such internal distribution cannot result in the absence of an externally identifiable responsible subject.

Accordingly:

Responsibility may be internally allocated, but it must be externally unified.

This distinction is fundamental.

• Internally: responsibility may be shared, segmented, or contractually redistributed.
• Externally: responsibility must remain attached to a clearly identifiable actor, capable of answering for the system as a whole.

Failing this, the system reverts to a state of organized irresponsibility.

IX. Limits and Risks of Purely Formal Attribution

However, the mere designation of a single entity as “responsible” is not sufficient if it is not accompanied by actual control, authority, and capacity to act.

A purely formal attribution, whereby one actor is presented as responsible while lacking effective control over the system, constitutes: a fictitious attribution of responsibility.

Such configurations must be rejected.

Responsibility cannot be reduced to:

• a contractual label,
• a commercial interface,
• or a communication strategy.

It must correspond to: effective power over the organization, deployment, and functioning of the system.

Otherwise, the appearance of responsibility conceals its absence.

 

Responsibility may be distributed in operation,
but it must be unified in attribution.

 

Conclusion

Artificial intelligence does not alter the nature of responsibility.
It reveals the deficiencies in how responsibility is structured and attributed.

The reliance on notions such as “distributed accountability” and the confusion between governance and responsibility obscure this reality.

The task is not to reinvent responsibility.

It is to reassert and reattach it.

Thus, the principle stands:

No intelligence can exist without responsibility.
The only question is where it must be attached.

 End

Patrick Houyoux, LL.M.
President — PT SYDECO

Sovereign Cybersecurity & Artificial Intelligence

“ARDENS DEFENDERE”