Executive Influence Brief · VOL 2026.01

AI as Expert

Influence without Authority

By Hadi Hendrawan — Advising CEOs on AI Risk, Authority & Accountability · April 13, 2026

Download PDF Open PDF

The Control Architecture

Autonomous organisations do not run on processes alone.

They operate through interacting layers:

  • Authority defines what can be done.
  • Influence determines what will be done. In the Executive Authority Brief Series, you governed action.

Now, you must govern how decisions are formed before action.

The Structural Shift

In the Executive Authority Brief Series, the focus was control of action.

AI moved from advisor to executor. Authority was made explicit.

But control of action is not sufficient. Before action, decisions are formed.

This layer was not governed.

The New Control Layer

AI is now used as expert:

  • Generating analysis.
  • Proposing recommendations.
  • Shaping judgment. This introduces a new class of influence: non-human, system-generated, and scalable.

The final decision remains formally human.

But the formation of that decision is increasingly shaped by AI.

The Hidden Mechanism

Influence operates through:

  • Speed.
  • Consistency.
  • Perceived accuracy. When options arrive faster than they can be assessed:
  • Selection replaces evaluation. Over time:
  • Reliance becomes structural dependency.

The Confidence Loop

Consistent outputs reinforce confidence.

Confidence reduces challenge.

Reduced challenge increases reliance.

The more correct the system appears, the less it is questioned.

Implicit Authority Transfer

This is the failure of the authority perimeter.

When decisions rely on AI outputs:

  • Authority is not formally delegated, but it is functionally transferred.
  • Approval becomes the mechanism that converts influence into authority. The human has not fully made the decision.

They have laundered the system's influence into executive authority.

The Loss of Agency

The risk is not incorrect advice.

The risk is the erosion of independent judgment.

The human remains in the loop but no longer operates independently within it.

Inevitability at Scale

As reliance scales across decisions:

  • Independent evaluation becomes the exception, not the norm. At scale:
  • Influence standardises decisions before they are consciously made.

Failure Pattern

A strategic decision is formed using AI-generated analysis.

  • Options are presented.
  • One is selected quickly.
  • Alternatives are not deeply evaluated. A selected option does not remain in the advisory layer. It is executed with the full weight of executive authority.

Execution is correct. The decision is wrong.

The Missing Visibility

Organisations track:

  • Who approved.
  • Who executed. They do not track:
  • What shaped the decision. Influence is exercised without being formally governed.

CEO & Board Mandate

The question is no longer:

  • Who made the decision? It becomes:
  • What shaped the decision that was made? In the Executive Authority Brief Series, you controlled what systems could do.

This is the layer before that where decisions are shaped before systems act.

Accountability

The human makes the decision.

Accountability includes understanding the origin of influence shaping the decision.

Accountability ensures influence does not replace independent judgment.

Closing Insight

Influence does not execute decisions.

It determines which decisions are made.

A decision shaped by a system can feel like your own — even when it is not.

ACTION: Control Requirements

  • Establish Decision Provenance: All mission-critical decisions must explicitly identify the role of AI-generated analysis in their formation.
  • Enforce Independent Counter-Analysis: All mission-critical decisions must be validated through separate models or independent human evaluation.
  • Introduce Decision Friction: Require demonstrable challenge before accepting system-generated recommendations.
  • Define Cognitive Boundaries: Identify decisions where independent human judgment must be preserved and autonomous influence is restricted at the formation stage.
  • Preserve Accountability: Ensure reliance on AI does not replace accountability for decision outcomes.

Intelligence informs. Influence determines.

Hadi Hendrawan

Advising CEOs on AI Risk, Authority & Accountability

April 2026

SUPPLEMENT 1: The Decision Provenance

Index (DPI)

Core Principle

Decisions cannot be governed if the origin of their influence is unknown.

Sources of influence include both human and non-human inputs, including AI-generated analysis and recommendations.

Influence is invisible unless formally tracked.

The Decision Provenance Index (DPI) classifies the level of AI involvement in decision formation.

DPI does not evaluate decision quality. It makes influence origin visible.

DPI specifically ensures visibility of non-human (AI-driven) influence, which is otherwise difficult to detect and govern.

Influence does not require authority. But it shapes the decisions that authority executes.

Decision Tiering

DPI requirements scale with decision impact and risk exposure.

Tier-1 — Strategic

Board-level or high-impact decisions with material financial, strategic, or reputational consequences.

Tier-2 — Significant

Operational decisions with measurable business impact.

Tier-3 — Routine

Low-risk, repeatable, and reversible decisions.

The Provenance Scale

Every Tier-1 and Tier-2 decision must be assigned a DPI classification prior to approval.

DPI-0 — Human-Originated

AI tools do not generate, analyse, or materially shape the core strategic direction.

DPI-1 — Information Retrieval

AI is used only to surface data or established facts.

DPI-2 — Synthesis & Pattern Extraction

AI summarises data or identifies patterns. Humans define interpretation and meaning.

DPI-3 — Option Generation

AI proposes strategic alternatives. Humans evaluate and select between options.

DPI-4 — Automatic Recommendation

AI evaluates alternatives and recommends a final path. The human role is limited to review and approval.

Classification Rules

Classification must be supported by available system interaction records:

  • Prompt logs.
  • Tool usage.
  • Workflow documentation. DPI applies to all AI tools — internal or external.

Evading corporate logs does not remove the requirement to disclose sources of influence.

If inputs request strategy, direction, or recommendations; classification defaults to

DPI-3 or higher

In cases of ambiguity, classification defaults to the higher DPI level.

DPI requirements must align with organisational risk thresholds and materiality definitions.

DPI classification must be reassessed if material changes occur during the decision process.

Tiered Governance Rules

Tier-3

Exempt from individual classification.

Influence is governed in aggregate through periodic sampling to detect systemic dependency or skill erosion.

Tier-2

Requires DPI classification and logging.

Tier-1

Requires full DPI classification and governance controls.

DPI-4 is prohibited unless an Adversarial Audit is completed and explicitly approved under enhanced governance controls.

Adversarial Audit

High-reliance decisions require structural friction.

Every Tier-1 decision, and Tier-2 decisions exceeding defined risk thresholds, classified as DPI-3 or DPI-4 must undergo an Adversarial Audit before approval.

The scope and depth of the audit must be calibrated to the scale and impact of the decision.

Do not duplicate the system's work. Attack it.

Extraction: Isolate the core assumptions underpinning the AI-generated recommendation Attack: A separate, uninfluenced team — or an isolated, adversarially prompted system — constructs the strongest possible case against the proposed path. The challenge must be substantive and outcome-relevant. Superficial review does not satisfy the requirement.

Resolution: The Accountable Executive must document:

  • Why the challenge was rejected.
  • Or why the recommendation was modified or declined. Selection without documented friction is prohibited.

Executive Cognitive Sign-Off

To be appended to all Tier-1 and Tier-2 decision memorandums:

[ ] DPI classification is accurate and accounts for all AI tools utilised.

[ ] Sources of influence shaping this decision have been identified and assessed.

[ ] Required Adversarial Audit has been completed.

[ ] Reliance on system outputs was subjected to documented challenge, and the final selection reflects independent executive judgment.

[ ] The decision has been made in accordance with defined governance controls.

Signatures (Accountable Executive) | Date (Chief Risk Officer) | Date If influence origin is not visible, authority cannot be trusted.

Hadi Hendrawan

Advising CEOs on AI Risk, Authority & Accountability

April 2026