The Challenge

AI didn’t just change what organisations can do. It changed who effectively exercises authority.

Organisations are deploying AI capability faster than they are governing the decisions that capability shapes. The result is a structural divergence between formal authority and effective control, a gap that existing governance frameworks are not designed to close.

01 How AI Changed Authority

AI has transformed the capability floor for every organisation simultaneously. But capability is not authority. As the floor rises for everyone, competitive advantage migrates upward, from what an organisation can do to what it chooses to do, and the quality of the systems through which those choices are made.

Every organisation with access to commercial AI infrastructure now has access to substantially similar capabilities: comparable language models, comparable reasoning systems, comparable analytical tools. The capability that once required significant investment and specialised expertise is now available as a commodity service.

This has a structural consequence. When the capability baseline rises for all competitors simultaneously, it ceases to be a source of advantage. An organisation that is excellent at deploying AI tools is not, by virtue of that excellence, better positioned than its competitors unless its competitors are less excellent. And as capability becomes a commodity, that differential shrinks.

The organisations that will consistently outperform are not those with better tools. They are those that exercise better authority with what they have.

But AI does not only raise the floor of what organisations can do. It reshapes the conditions under which decisions are made. When AI systems filter information, generate options, rank possibilities, and shape the range of choices available before the human decision maker acts, they are not merely enabling decisions. They are participating in them. And that participation has governance implications that most organisations are not yet designed to address.

01.1

Commodity capability

Commercial AI infrastructure is available to all organisations at equivalent cost. The capability gap that once separated leaders from followers has closed to near zero at the baseline level. Capability is no longer the source of lasting advantage.

01.2

Participation before decision

AI participation in consequential decisions operates primarily through the possibility space, the set of options available to the decision maker before the decision is made. Governance that does not see this layer is not governing the decision.

01.3

Advantage migration

As AI raises the capability floor, competitive advantage migrates upward from what an organisation can do to what it chooses to do, and the quality of the decision systems through which those choices are constitutionally made and governed.

02 The Decision Boundary

There is a line in every consequential decision where AI influence ends and human authority must begin. That line, the Decision Boundary, is the central governance object of the AI era. Most organisations cannot locate it. Most governance frameworks do not look for it.

The Decision Boundary is not a technical concept. It is a constitutional one. It describes the threshold at which responsibility for a consequential decision shifts from the systems that shaped it to the person or body formally designated to make it. Where the Decision Boundary sits, and whether it is designed or merely assumed, determines whether an organisation's decision system can be governed at all.

In a world without AI, the Decision Boundary was relatively legible. A human decision maker encountered a situation, formed a view, and made a choice. The conditions of that choice, what information was available, what options were visible, what constraints were operative, were generally traceable. Authority and effective control were roughly aligned.

AI disrupts this alignment. When AI systems have already filtered the information, constructed the options, weighted the alternatives, and shaped the framing of the decision before the human authority holder acts, the effective locus of decision has migrated. The formal decision, the human's choice, rests on a foundation that the formal authority holder may not have seen, examined, or authorised.

Learning is adaptive. Judgement is selective. AI raises the floor of organisational learning for everyone. It does not raise the quality of organisational judgement, which is determined by the systems through which learning is converted into decisions.

The Decision Boundary concept makes a precise claim: the question for governance is not whether AI participates in decisions. In most organisations, it already does. The question is whether that participation is constitutionally specified, visible to those who are accountable, and designed to support rather than supplant human authority. Where the boundary is undesigned, authority has not been eliminated. It has been obscured.

02.1

The boundary is constitutional

The Decision Boundary defines who bears final authority, and therefore accountability, for a consequential choice. This is a constitutional question, not a technical one. It must be specified deliberately or it will be occupied by default.

02.2

AI operates before the decision

AI influence is primarily exercised in the possibility space, before the human decision maker acts. Governance that begins at the point of human decision is governance that arrives too late to see where effective authority has already operated.

02.3

Obscured is not eliminated

When the Decision Boundary is undesigned, authority is not eliminated. It migrates to the system that most effectively shapes the possibility space. Without specification, that system is AI, not by design, but by default.

03 Authority Drift

Without constitutional specification, authority drifts. It drifts from formal holders to those who effectively shape what is decided. In organisations deploying AI without governance design, it drifts systematically, and predictably, toward AI systems.

Authority drift is not an event. It is a trajectory. It does not happen because anyone decides to transfer authority to AI. It happens because organisations invest in AI capability without specifying how that capability interacts with their constitutional authority structure. Each individual investment appears routine. The cumulative effect is a progressive migration of effective decisional control.

The drift mechanism operates through three channels. First, information filtering: as AI systems determine what information reaches decision makers, they shape what is cognised, what is excluded, and what range of options appears normal. Second, option generation: as AI systems produce the choices from which decision makers select, they determine what is visible and what is not. Third, framing: as AI systems present decisions in particular ways, they shape the weight that decision makers assign to competing considerations.

None of these channels requires that AI be given formal authority. They operate through the possibility space, not the formal decision. The formal decision maker retains nominal authority. But the effective conditions of that authority have been redesigned, without constitutional specification of what that redesign means for governance.

03.1

Drift is structural

Authority drift is not exceptional. It is a structural prediction of what happens when AI capability is deployed without constitutional specification. The conditions that produce it are present in most organisations investing in AI at scale.

03.2

Investment decisions enable drift

Each AI investment that shapes the possibility space of consequential decisions, without governance specification, is a drift event. Individually, each appears routine. Cumulatively, they represent a constitutional change that no one authorised.

03.3

Nominal versus operative authority

The Constitutional Gap is the divergence between who formally holds authority and who effectively exercises it. Authority drift widens this gap. An unmeasured, unmanaged drift rate is the primary indicator that an organisation's decision system is ungoverned.

04 The Accountability Gap

When AI shapes the conditions of a consequential decision, accountability for that decision becomes structurally ambiguous. The formal decision maker is accountable for the choice they made, but not for the possibility space from which they chose. That is where accountability stops. And that is where AI operates.

Accountability requires a traceable chain from decision to decision maker. In AI augmented organisations, that chain is systematically disrupted at the possibility space, the layer through which AI influence operates. The decision maker can account for their choice. They cannot generally account for the conditions under which that choice was constituted.

This is not a failure of individual accountability. It is a structural property of a decision system without constitutional specification. Where the conditions of a decision are not visible to the authority holder, where the possibility space is opaque, accountability is reduced to accountability for the output of a system the accountable party cannot see.

An organisation that cannot trace where AI influence ended and human authority began cannot assign accountability for the decisions that sit at that boundary. Most consequential decisions in AI augmented organisations sit precisely there.

The accountability gap does not only affect internal governance. It affects stakeholder trust, regulatory compliance, and the capacity of boards and executive teams to give an account of their organisation's most consequential choices. Governance bodies that claim oversight of AI augmented decisions, without access to the decision record, the possibility space, and the conditions under which the boundary was drawn, are asserting governance, not exercising it.

04.1

Accountability stops at the possibility space

Formal accountability covers the decision. It does not cover the conditions under which the decision was constituted. AI operates primarily in those conditions. The accountability gap is exactly as wide as the unspecified possibility space.

04.2

Governance without a subject

Governance bodies that cannot see the decision record, what was decided, by whom, from what possibility space, under what AI participation, are not governing the decision system. They are governing a description of one that may no longer correspond to the one operating.

04.3

Legitimacy depends on traceability

Decision legitimacy requires that the decision was made by a recognised authority through an adequate process. Where the process is opaque, where AI participation is unspecified, the legitimacy of the decision is structurally compromised, regardless of the quality of the outcome.

05 Existing Governance Limitations

The major governance disciplines each address a genuine problem. Each assumes an answer to the constitutional question, who has authority over the decisions this framework governs, that none of them supplies. That assumption is the gap.

Enterprise Architecture governs the technology infrastructure through which decisions are enabled. Organisation Design governs the reporting structures through which authority is formally held. AI Governance governs the risk properties of AI systems and their outputs. GRC governs compliance with policies and regulations. Process Engineering governs the workflows through which decisions are executed.

Each of these disciplines is designed to govern something real and important. And each was designed for organisations in which decisions are made by human authority holders, not for organisations in which AI participates in shaping the possibility space before the decision is made.

The constitutional layer is not missing because governance practitioners are careless. It is missing because no existing discipline has claimed responsibility for it. Each framework assumes the answer. None provides it.

The gap is not in any individual discipline. It is in the space between them. Every framework assumes that someone, somewhere, has specified who has authority over the decisions it governs. No framework asks who that is, checks whether it is still true, or measures whether the operative authority structure still corresponds to the constitutional specification. That is the work of Decision Systems Engineering.

Discipline
Governs
Assumes (but does not specify)
Enterprise Architecture
Technology systems enabling decisions
That authority holders exist with specified decision rights over what the technology enables
Organisation Design
Reporting structures and formal role responsibilities
That roles carry specified decision rights over categories of consequential choice
AI Governance
Risk properties of AI models and compliance of outputs
That human authority holders exist with power and accountability to act on AI outputs
GRC
Compliance with policies, regulations, and procedural requirements
That the policies enforced rest on a constitutional foundation specifying legitimate authority
Process Engineering
Workflows through which decisions are implemented
That the decisions those processes implement were made by legitimate authority holders
Decision Systems Engineering
The conditions under which consequential authority is exercised, observed, evaluated, and made to answer
Nothing. The Institute proposes it provides what the other disciplines assume.

06 Why Not Just AI Governance

AI Governance addresses a real problem: ensuring that AI systems are safe, compliant, and aligned with organisational policy. But it is not the same problem as decision governance. Confusing the two leaves the constitutional gap open.

AI Governance asks: is this AI system operating safely? Does its output comply with applicable policies? Are its risks being managed? These are legitimate and important questions. Every organisation deploying AI should be asking them. But they are questions about the AI system, not about the decision system in which it participates.

The constitutional governance question is different: who has authority over the consequential decisions that AI participates in? From what possibility space are those decisions being made? Is that possibility space visible to those who are formally accountable? Can the exercise of authority be traced, evaluated, and made to answer? These are questions about the decision, not the AI.

An organisation can have excellent AI Governance, rigorous model risk management, comprehensive output monitoring, full regulatory compliance, and still have no governance over the decisions that AI shapes. It can know everything about its AI systems and nothing about whether its decision system is still being governed by the people formally designated to govern it.

AI governance asks whether the system is safe. Decision governance asks whether the authority is intact. Both questions matter. Only one of them is currently being asked at scale.

The problem is not that AI Governance exists. It is that organisations treat it as sufficient when it is not designed to be. AI Governance governs the tool. Decision Systems Engineering governs the conditions under which authority is exercised through that tool. Both are necessary. Neither substitutes for the other.

06.1

Different objects

AI Governance governs AI systems. Decision governance governs the conditions under which authority is exercised. These are distinct objects requiring distinct disciplines. Conflating them leaves the constitutional question unanswered.

06.2

Safe AI, ungoverned decisions

An organisation with rigorous AI Governance can simultaneously have no governance over its decision system. Safe, compliant AI can operate within a constitutional vacuum, shaping consequential decisions without any specification of how that shaping is authorised or overseen.

06.3

The constitutional foundation

What AI Governance assumes, that human authority holders exist, are specified, and have decision rights over what AI produces, is precisely what Decision Systems Engineering proposes to provide. DSE is not a competitor to AI Governance. The Institute proposes it as its constitutional foundation.

Decision Systems Engineering Research Programme