How AI Enabled Systems Reshape Authority in Public Institutions

Governments are moving rapidly to embed AI-enabled systems into frontline administration. Yet a central question remains unresolved: where is public authority actually being exercised inside these systems, and how can institutions govern it?

Two new preprints released this month address this problem from different angles. Together with earlier work on Regulatory Dark Matter, they outline a governance architecture for understanding how authority is constructed, distributed, and operationalised in contemporary public administration.

The first paper examines governmentality, fragmentation, and visibility in intelligence-led regulation. The second introduces a structural model of execution-time authority in AI-enabled decision systems. Both extend a foundation established in earlier work on Regulatory Dark Matter, which described the invisible forces that shape regulatory behaviour beyond formal rules.

1. Regulatory Dark Matter as the upstream problem

Regulatory Dark Matter refers to the informal rules, tacit practices, analytic habits, and institutional blind spots that shape regulatory behaviour but rarely appear in legislation or policy. It explains why regulators act the way they do, even when formal rules suggest otherwise.

This concept provides the upstream frame for understanding how authority is produced before it ever reaches a decision system. It shows that authority is not only delegated through statutes and policies. It is also constructed through the invisible structures that define what regulators see, prioritise, and respond to.

The two new preprints extend this foundation by examining how these invisible forces operate in intelligence-led regulation and how they manifest inside AI-enabled administrative systems at the point where authority is actually exercised.

2. Governmentality, Fragmentation, and Visibility: Rethinking Intelligence-Led Regulation and Informal Rules in Australia

Preprint: Governmentality, Fragmentation, and Visibility: Rethinking Intelligence-Led Regulation and Informal Rules in Australia

DOI: 10.2139/ssrn.6671179

This paper argues that intelligence-led regulation is not simply a technical practice but a form of governmentality. It is a way of seeing, knowing, and acting that structures regulatory attention. It shows how:

• visibility architectures determine what becomes governable
• fragmentation creates uneven sightlines across institutions
• analytic practices shape the construction of risk
• informal rules and tacit heuristics drive regulatory behaviour.

The result is a regulatory environment where authority is produced through epistemic infrastructures, not just formal mandates. Intelligence systems do not merely inform decisions. They shape the conditions under which decisions are possible.

This provides the middle layer of the governance architecture: how authority is constructed and channelled before it reaches operational systems.

3. Execution Time Authority in AI-Enabled Administration

Preprint: Execution-Time Authority in AI-Enabled Administration: A Governance Architecture for Delegated Power

DOI: 10.5281/zenodo.20090096

This second paper examines what happens when authority moves into the execution time layer. This is the moment-to-moment operational space where automated systems make decisions faster than institutions can respond.

It introduces a structural model showing how authority migrates:

• from design time (legislation, policy, delegation)
• to run time (workflows, business rules, parameters)
• to execution time (models, optimisers, automated triggers).

When systems begin exercising unmandated authority at execution time, institutions lose control without noticing. This is not a malfunction. It is a structural property of AI-enabled administration.

This paper provides the downstream layer of the governance architecture, where authority is actually exercised inside modern decision systems.

4. A unified governance architecture

Taken together, the three works form a coherent architecture:

• Regulatory Dark Matter explains how authority is produced through invisible institutional forces.
• Governmentality, Fragmentation, and Visibility explains how authority is constructed and channelled through epistemic infrastructures.
• Execution Time Authority explains how authority is performed inside AI-enabled systems at the point of decision.

This architecture shows that authority in contemporary public administration is not static. It is produced upstream, shaped midstream, and exercised downstream, often in places institutions cannot see.

5. Why this matters for public institutions now

For public institutions, these are not abstract questions. AI-enabled systems are already:

• filtering which cases are seen and which are not
• shaping how risk is defined and prioritised
• constraining what frontline staff can decide
• generating outcomes that appear neutral but embed institutional blind spots

In this environment, authority can shift from legislated and delegated structures to analytic and execution-time infrastructures. When that happens, accountability frameworks, integrity systems, and traditional oversight mechanisms are all working on the wrong layer.

The risk is not only unfair or harmful outcomes. The deeper risk is that institutions no longer know where their own authority is being exercised.

6. Using the governance architecture as a practical tool

The value of this governance architecture is practical rather than theoretical. It gives executives, central agencies, and integrity bodies a way to ask concrete questions about AI-enabled systems, such as:

• Where is authority being produced before it reaches this system?
• Which epistemic and intelligence structures are shaping what the system can see?
• At what points does the system exercise execution time authority rather than simply implementing a mandate?
• Which parts of this authority are visible to existing oversight mechanisms, and which are not?

By treating Regulatory Dark Matter, governmentality, and execution time authority as linked components of a single architecture, institutions can begin to design governance that matches the systems they now rely on.