An AI system reviewing a portfolio identifies a project that is deteriorating faster than its status reporting suggests. The evidence is credible, the exposure is material, and the recommendation is clear: intervene now.

The PMO can raise the issue, the sponsor owns the initiative, and a governance forum can review it. An executive may ultimately decide what happens next.

But who can actually change the course of the project today?

This is a problem organizations will encounter more frequently as AI improves their ability to detect risk, challenge established narratives, and recommend action. Better intelligence can make a problem visible much earlier without changing the authority available to respond to it.

An organization can therefore improve sensing, analysis, and recommendation substantially while leaving the underlying decision rights untouched. The result is a gap between what the organization knows and what it is authorized to do about it. AI does not create that gap, but improved intelligence can make it much harder to ignore.

1. Better intelligence exposes existing decision rights

Many organizations already separate the ability to identify a problem from the authority to determine its consequence.

A PMO may report that a program is unlikely to achieve its objectives but lack the authority to stop or materially redirect it. A risk function may identify exposure while the relevant business owner retains the decision. An operations group may recognize that a customer commitment is becoming untenable while commercial authority sits elsewhere.

Those arrangements are not inherently wrong. Authority is distributed for legitimate reasons, including accountability, expertise, financial responsibility, regulatory requirements, and organizational design.

The difficulty appears when the management system is unclear about what should happen once credible intelligence crosses an important threshold.

Where signals emerge gradually through periodic reporting and established review cycles, ambiguity in decision rights can remain less visible for longer. A concern enters a status report, is discussed in a meeting, and moves through an escalation route before a consequential decision becomes unavoidable.

AI can compress that sequence.

A material signal can be detected quickly, relevant evidence can be assembled much earlier, and a recommendation can arrive before the organization has activated the management process through which someone is legitimately able to respond.

The analytical case may become compelling before the authority question has been resolved.

2. A recommendation is not a decision

AI systems can produce recommendations that are increasingly specific, timely, and supported by substantial evidence. The recommendation might be to stop an activity, reallocate capacity, increase a reserve, change a supplier, escalate a customer issue, suspend a release, or investigate a control failure.

The precision of the recommendation can create the impression that the management problem has also been resolved.

It has not.

A recommendation is an input to management. A decision creates organizational commitment. Authority determines who may legitimately make that commitment, and execution determines whether the decision can actually change what happens.

Those distinctions matter even when technology compresses the time between them.

Consider a portfolio in which AI identifies a sponsor-backed initiative as a likely source of significant future loss. The PMO may have the evidence and may strongly recommend intervention. If the PMO's mandate is limited to reporting and escalation, however, the recommendation does not acquire additional authority because it was generated or supported by AI.

The same principle applies in operational settings. An AI-enabled system may detect that an established threshold has been crossed, but whether it can change a workflow, restrict access, halt a transaction, or redirect resources depends on authority that must be deliberately assigned.

Capability and authority are different organizational properties.

Authority and accountability should not be collapsed either. Authority concerns the legitimate right to decide or act. Accountability concerns who or which governance role remains answerable for the relevant decision, execution, or resulting organizational responsibility. AI may participate in a decision process or act within delegated boundaries, but the applicable organizational accountabilities must remain explicitly assigned.

As AI becomes more capable, those distinctions become more important, not less.

3. Authority has to remain usable when the decision arrives

Formal decision rights are necessary, but they are not sufficient.

An organization may be able to point to a policy, governance matrix, or organizational chart that identifies the nominal decision owner. That does not necessarily mean the required authority is usable when the situation demands it.

The authorized person may be unavailable. The escalation path may require several intermediary steps. The decision may cross organizational boundaries that no single owner controls. A manager may technically possess authority but lack sufficient context, evidence, organizational backing, or practical access to exercise it when required.

In those situations, authority exists formally but is weak operationally.

This becomes more consequential when AI shortens the time between detection and the point at which intervention could still protect value. A management system that can identify deterioration within minutes but requires several days to locate and activate legitimate authority has improved intelligence without improving its capacity to respond.

The relevant management question is therefore broader than “Who owns the decision?”

Management needs more than a named owner. Authority must be clear for the decision, available when needed, and bounded so that people know what can be decided within existing limits. Where those limits are exceeded, there must be a workable escalation route and sufficient evidence for responsible exercise of the resulting authority.

The objective is not to push every decision upward or to give operating teams unlimited discretion. It is to ensure that legitimate authority can be reached and exercised at the level appropriate to the consequence of the situation.

4. AI delegation makes the boundary more important

The authority question becomes more demanding when AI moves beyond recommendation and begins to create consequence directly.

An AI-enabled system may be permitted to approve a low-value transaction, change a schedule, reroute work, modify a record, allocate capacity, or trigger another workflow. At that point, the organization has chosen to embed delegated decision or action authority within the operating design.

That delegation should be explicit.

The important questions are not only what the AI can technically do, but what it is legitimately permitted to decide or execute, under what conditions, within which boundaries, and with what route to human intervention when those conditions no longer hold.

Routine, low-consequence, reversible actions may be suitable for broader automation within clearly defined limits. Greater exposure, uncertainty, irreversibility, conflicting evidence, or a breach of an established boundary may require escalation or direct human authorization.

Delegation can otherwise shift incrementally. A system may begin by recommending an action, then prepare it automatically, and later execute it unless someone intervenes. The formal governance description may still imply a human decision while the operating reality has shifted part of the decision and action authority into the system.

That shift may be appropriate, but it should be intentional.

When AI moves from recommendation to consequence, the authority question becomes part of system design.

5. From insight to legitimate action

As AI improves management intelligence, organizations need to examine the route from insight to consequence with the same care they apply to the quality of the analytical capability itself.

For consequential decisions, management should be able to answer a small set of practical questions:

• What decision is actually being made?

• Who has legitimate authority to make the decision, and who has authority to execute or intervene?

• What may be decided within existing boundaries, and what requires escalation?

• Can that authority be exercised quickly enough for intervention still to matter?

• If AI is permitted to act, what decision or action authority has been delegated, and under what conditions?

• Where do the relevant decision and execution accountabilities remain?

These questions also show why “human in the loop” is often an incomplete description of governance. A human can be present without possessing meaningful authority. A nominal approval step can exist without giving the person sufficient time, context, or practical ability to exercise judgment. An escalation route can exist while still being too slow to influence the outcome.

The management requirement is not simply human presence. It is usable, legitimate authority at the point where consequence can still be shaped.

Organizations sometimes interpret reluctance to act on AI recommendations as a trust problem. Sometimes it is. But a strong recommendation may also expose an authority problem: acting on it requires someone to exercise decision or intervention authority that the management system has never clearly assigned or made usable.

Better intelligence can expose that gap, but greater analytical accuracy cannot resolve it. As AI closes the distance between signal, recommendation, decision, and possible action, organizations have to become more explicit about where decisions are made, how authority is exercised, how escalation and intervention work, and where accountability remains.

Insight without usable authority does not create action. It creates a management gap.