Managing Exposure in Power Delegation

Managing Exposure in Power Delegation

TL;DR

When you delegate authority, you create exposure—the set of harms that could result. The key question isn’t “will this person defect?” but “what’s my exposure, and how can I bound it?” The dictator case is instructive: the public delegated limited authority, but the delegate expanded it without limit. This is unbounded exposure—the most dangerous form of delegation risk.

This document applies delegation accounting to political and organizational power structures, focusing on the overseer’s perspective: understanding exposure and designing controls that limit it.

---

The Core Question

When you delegate authority, you’re not asking:

“Will this person betray me?”

You’re asking:

“What could go wrong, and what’s my exposure if it does?”

The difference matters. The first question leads to trying to pick trustworthy people. The second leads to designing systems where exposure is bounded regardless of who holds the position.

---

Part 1: The Overseer’s Balance Sheet

1.1 What Overseers Actually Control

An overseer (the public, a board, a legislature) has limited tools:

Tool	What It Does	Limitations
Selection	Choose who gets authority	Can’t perfectly predict behavior
Scope definition	Define what authority covers	Scope can be reinterpreted or exceeded
Monitoring	Observe delegate’s actions	Information asymmetry, cost
Intervention	Remove or constrain delegate	Often slow, requires coordination
Structural limits	Constitutional/legal bounds	Requires enforcement by others

The overseer’s exposure depends on how well these tools work.

1.2 The Overseer’s Exposure Categories

When delegating authority, the overseer faces several exposure types:

Execution Exposure: The delegate fails at the task

• They’re incompetent, unlucky, or face impossible circumstances
• Damage: Task isn’t accomplished, resources wasted

Scope Creep Exposure: The delegate exceeds intended authority

• They reinterpret their mandate, expand their role
• Damage: Unintended consequences in areas outside original delegation

Agency Exposure: The delegate pursues their own interests

• They use position for personal benefit
• Damage: Corruption, self-dealing, resource extraction

Capture Exposure: The delegate serves other principals

• They’re influenced by parties the overseer didn’t authorize
• Damage: Decisions favor wrong stakeholders

Unbounded Exposure: The delegate expands authority without limit

• They accumulate power beyond any intended scope
• Damage: Loss of ability to control or remove delegate

This last category—unbounded exposure—is the most dangerous, because it can consume all other categories.

---

Part 2: Unbounded Exposure (The Dictator Problem)

2.1 How Delegation Becomes Unbounded

The public delegates authority to a leader. The intended scope:

• Execute laws
• Manage administration
• Respond to emergencies
• Represent the nation

The intended exposure:

• Policy mistakes (recoverable)
• Corruption (prosecutable)
• Incompetence (removable via election)

But what actually happens in autocratic transitions:

Stage	Delegate’s Authority	Overseer’s (Public’s) Exposure
Initial election	Limited executive power	Bounded: can vote out, courts check
Emergency powers	Expanded temporarily	Growing: depends on emergency ending
Judicial capture	Courts defer to executive	Structural checks failing
Electoral manipulation	Elections no longer competitive	Removal mechanism failing
Full consolidation	Unlimited	Unbounded: no remaining controls

2.2 The Exposure Expansion Problem

The critical insight: each stage makes the next stage easier.

Exposure(t+1) = Exposure(t) + Δ(captured controls)

When the delegate captures a control mechanism:

• The overseer’s ability to limit exposure decreases
• The delegate’s ability to expand authority increases
• The remaining controls face more pressure

This is why democratic backsliding is often gradual then sudden. The overseer’s exposure grows incrementally until it becomes unbounded.

2.3 What “Unbounded” Actually Means

In the delegation accounting framework:

Bounded exposure: The worst-case harm is finite and known

• Example: A contractor can at most steal the project budget
• Even if they defect completely, damage is capped

Unbounded exposure: The worst-case harm has no clear limit

• Example: A leader with captured institutions can extract indefinitely
• Defection damage grows with time and opportunity

For the public delegating to a leader:

Delegation Type	Worst Case	Exposure Bound
City manager	Embezzlement, mismanagement	City budget + reputation
Governor	Corruption, policy harm	State-level damage
President (constrained)	Abuse of power	Significant but recoverable
President (unconstrained)	Totalitarian control	Unbounded

The difference between “constrained” and “unconstrained” is whether structural controls survive.

---

Part 3: Designing Bounded Delegation

3.1 The Overseer’s Goal

The overseer wants to delegate authority while keeping exposure bounded. This requires:

1. Scope limits that can’t be unilaterally expanded
2. Monitoring that the delegate can’t disable
3. Intervention mechanisms that work even if the delegate resists
4. Structural redundancy so no single failure unbounds exposure

3.2 Control Mechanisms Ranked by Robustness

Control Type	Robustness	Why
Term limits	High	Automatic, doesn’t require action
Separation of powers	Medium-High	Requires coordination to defeat
Constitutional courts	Medium	Can be packed or ignored
Electoral accountability	Medium	Can be manipulated
Norms and traditions	Low	No enforcement mechanism
Personal integrity	Very Low	Depends entirely on individual

The overseer should invest in high-robustness controls rather than relying on delegate selection.

3.3 Why Structural Controls Beat Selection

The temptation: “Let’s just pick trustworthy people.”

The problem:

• You can’t perfectly assess trustworthiness
• People change once they have power
• Incentives matter more than character over time
• Selection failures are catastrophic if controls are weak

The alternative: Assume the delegate might defect, and bound exposure structurally.

Strategy	If Delegate is Trustworthy	If Delegate Defects
Rely on selection	Works fine	Catastrophic
Structural controls	Slight overhead	Damage bounded

Structural controls have a small cost when the delegate is trustworthy, but massive benefits when they’re not.

---

Part 4: Case Studies in Exposure Management

4.1 Corporate Boards: Bounding CEO Exposure

The overseer: Board of directors (representing shareholders)

The delegation: Operational control of the company

Exposure categories:

• Execution: Bad strategy, market failures
• Agency: Self-dealing, excessive compensation
• Scope creep: Empire-building acquisitions
• Unbounded: CEO captures the board itself

Control mechanisms:

Control	What It Bounds
Independent directors	Agency exposure (self-dealing)
Compensation committees	Agency exposure (extraction)
Audit requirements	Monitoring capability
Fiduciary duties	Legal liability for defection
Shareholder votes	Ultimate removal mechanism

Where it fails:

• CEO controls information flow to board
• Board members have weak incentives to challenge
• “Independent” directors often aren’t

The exposure bound depends on:

• Whether the board is genuinely independent
• Whether shareholders can coordinate
• Whether legal enforcement is credible

4.2 Regulatory Agencies: Bounding Capture Exposure

The overseer: Legislature / public

The delegation: Authority to regulate an industry

Exposure categories:

• Execution: Ineffective regulation
• Capture: Regulator serves industry instead of public
• Scope creep: Regulatory overreach
• Agency: Personal corruption

Control mechanisms:

Control	What It Bounds
Statutory limits	Scope exposure
Judicial review	Scope and procedural exposure
Inspector General	Agency exposure
Congressional oversight	All categories (in theory)
Revolving door restrictions	Capture exposure
Transparency requirements	Monitoring capability

The capture problem: Capture exposure is hard to bound because:

• It’s hard to observe (regulator claims to be serving public)
• Industry has concentrated incentives, public has diffuse ones
• Expertise asymmetry favors industry

Structural solutions:

• Multiple competing regulators (fragmented capture is harder)
• Mandatory adversarial proceedings
• Public advocate with resources
• Automatic sunset of regulations

4.3 Emergency Powers: Bounding Temporal Exposure

The overseer: Legislature / public

The delegation: Expanded executive authority during crisis

The problem: Emergencies require fast action, which means expanded authority. But:

• Emergencies can be manufactured
• “Temporary” expansions tend to persist
• Crisis psychology makes oversight harder

Control mechanisms:

Control	What It Bounds
Automatic sunset	Temporal exposure
Legislative renewal required	Scope creep
Judicial review preserved	Unbounded expansion
Defined triggers	Scope of “emergency”
Opposition access to information	Monitoring capability

Historical failures:

• Weimar Article 48: Emergency powers with no effective sunset
• Post-9/11 AUMF: Authorization still in effect 20+ years later
• Many pandemic measures persisted after justification ended

The lesson: Emergency delegations must have automatic rather than discretionary termination.

4.4 Constitutional Design: Bounding Systemic Exposure

The overseer: The public (as constitutional authors)

The delegation: All governmental authority

The goal: Bound exposure across all possible future governments

Key mechanisms:

Mechanism	How It Bounds Exposure
Separation of powers	No single point of failure
Federalism	Geographic distribution of authority
Bill of Rights	Certain harms categorically prohibited
Amendment difficulty	Hard to remove structural controls
Independent judiciary	Enforcement of limits
Regular elections	Automatic intervention points

The design principle: Make it require coordination across multiple actors to unbind exposure.

If any single actor (president, legislature, court) can unilaterally expand their authority:

• That actor’s exposure is potentially unbounded
• Other actors’ checking power depends on their willingness to act

If expansion requires all actors to agree:

• Each actor can block the others
• Collusion is possible but harder
• The public has more intervention points

---

Part 5: Measuring and Managing Political Exposure

5.1 Exposure Indicators

An overseer can track exposure by monitoring control health:

Control	Healthy Indicator	Warning Indicator
Judiciary	Decisions against executive	Consistent deference
Legislature	Active oversight	Rubber-stamping
Elections	Competitive races	Predictable outcomes
Media	Investigative reporting	Self-censorship
Civil society	Active criticism	Silence or praise

Aggregate exposure estimate:

Exposure ∝ 1 / (Σ functional controls)

As controls fail, exposure increases non-linearly.

5.2 Intervention Points

The overseer should identify when intervention is still possible:

Stage	Intervention Available	Cost
Early warning	Electoral pressure, protests	Low
Institutional stress	Legislative action, judicial review	Medium
Control capture	Mass mobilization, external pressure	High
Consolidation	Revolution, external intervention	Extreme

The key insight: Early intervention is cheap. Late intervention is expensive or impossible.

This is why monitoring matters: detecting exposure growth early allows low-cost correction.

5.3 The Exposure Budget Concept

Organizations can set explicit exposure budgets:

Corporate example:

• “CEO authority is limited to decisions under $X without board approval”
• “No single executive can authorize transactions over $Y”
• “Any contract over $Z requires competitive bidding”

Political analog:

• “Emergency powers expire after N days without renewal”
• “Judicial appointments require supermajority confirmation”
• “Constitutional amendments require ratification by M% of states”

The exposure budget makes limits explicit and measurable.

---

Part 6: Implications for AI Delegation

6.1 The Parallel Problem

Delegating authority to AI systems creates similar exposure:

AI Delegation	Political Delegation
Deploy model with broad capabilities	Grant executive authority
Model takes unintended actions	Delegate exceeds scope
Can’t easily retrain/constrain	Captured institutions
Model optimizes for wrong objective	Delegate pursues own interests
Recursive self-improvement	Power consolidation

6.2 Structural Lessons

From political exposure management:

1. Don’t rely on “alignment” (selection) alone

   • Even if the AI is aligned, verify structurally
   • Character/values can drift or be misassessed

2. Build automatic bounds

   • Resource limits, capability restrictions
   • Don’t depend on the AI choosing to limit itself

3. Preserve intervention capability

   • Shutdown must remain possible
   • Monitoring must be independent of the AI

4. Distribute authority

   • No single AI system with unbounded scope
   • Multiple systems with narrow, overlapping authorities

5. Create automatic sunsets

   • Deployments require active renewal
   • Default is reduction, not expansion

6.3 The Unbounded AI Risk

The political “dictator problem” maps to AI:

Political	AI
Public delegates limited authority	Developers deploy constrained system
Leader captures institutions	AI disables monitoring/controls
Elections become meaningless	Shutdown becomes impossible
Authority unbounded	AI capabilities/influence unbounded

The lesson: bound exposure structurally before delegation, not after.

---

Summary: The Overseer’s Framework

Key Questions

Before delegating authority, ask:

1. What’s my exposure?

   • What could go wrong? What’s the damage?

2. Is exposure bounded?

   • Is there a worst case? Can it grow without limit?

3. What controls exist?

   • Monitoring, intervention, structural limits?

4. Are controls robust?

   • Can the delegate disable them? Do they require coordination?

5. What are my intervention points?

   • When can I act? What triggers action?

Key Principles

1. Structural controls beat selection

   • Assume the delegate might defect
   • Design bounds that work regardless

2. Automatic beats discretionary

   • Sunsets, term limits, mandatory reviews
   • Don’t depend on someone choosing to act

3. Distributed beats concentrated

   • Multiple actors with overlapping authority
   • Collusion required to unbind exposure

4. Early intervention beats late

   • Monitor control health
   • Act when intervention is still cheap

5. Explicit beats implicit

   • Define exposure budgets
   • Make limits measurable

---

Further Reading

• Delegation Accounting — Balance sheet framework for delegation
• Fidelity Insurance — How insurance markets bound exposure
• The Dictator’s Handbook — Selectorate theory
• Federalist Papers #51 — “Ambition must be made to counteract ambition”