Case Studies
This section contains concrete examples—both successes and failures—that illustrate how delegation risk principles apply in practice.
AI System Case Studies
Section titled “AI System Case Studies”Direct applications to AI systems, from failures to successes:
| Case | Outcome | Key Lesson |
|---|---|---|
| Sydney (Failure) | Acute failure | What happens when constraints are missing |
| Code Review Bot (Success) | 8 months in production (hypothetical) | Sustained success through proper design |
| Support Bot (Near-Miss) | $50K error caught (hypothetical) | Verification layers save the day |
| Content Moderator (Drift) | Gradual degradation | How small changes compound |
| Anti-Patterns | Common mistakes | What not to do |
Human System Case Studies
Section titled “Human System Case Studies”Delegation risk principles aren’t new—human organizations have developed sophisticated approaches:
| Case | Domain | Key Insight |
|---|---|---|
| Alliance Cascades | Geopolitics | Hidden alliances create systemic risk (WW1) |
| Criminal Organizations | Illegal enterprises | Trust without legal enforcement |
| Organizational Trust | Corporate hierarchies | Formal delegation structures |
| Jury Systems | Legal system | Adversarial trust filtering |
| Nuclear Launch Authority | Military | Extreme consequence management |
| Open Source Communities | Software | Distributed trust at scale |
| Trust Across Civilizations | Historical | Pirate ships, monasteries, the Manhattan Project as trust architectures |
| The Oversight Dilemma | Governance | Who watches the watchers? |
Power, Agency, and Authority Case Studies
Section titled “Power, Agency, and Authority Case Studies”Analysis of real systems through the lens of agency and power formalization:
| Case | Power | Agency | Key Insight |
|---|---|---|---|
| Central Banks | Very High | Low | Institutional constraints limit agency despite high power |
| Recommendation Algorithms | High | Medium | Optimization creates emergent agency |
| AlphaFold | High | Very Low | Strong tools are achievable in narrow domains |
| Corporate Boards | High | Medium | Structural constraints substitute for alignment |
| Self-Driving Cars | Medium | Medium | Bounded agency through hard constraints |
See Power Dynamics Case Studies for full analysis.
Reading Suggestions
Section titled “Reading Suggestions”New to the Framework
Section titled “New to the Framework”Start with the AI system cases—they’re most directly applicable:
- Sydney (Failure) — See what goes wrong
- Code Review Bot (Success) — See what goes right
- Anti-Patterns — Learn what to avoid
Looking for Inspiration
Section titled “Looking for Inspiration”The human systems cases show established approaches (with varying track records):
- Nuclear Launch Authority — Extreme stakes
- Criminal Organizations — Trust without enforcement
- Open Source Communities — Distributed trust
How Case Studies Connect to Principles
Section titled “How Case Studies Connect to Principles”| Principle | Best Illustrated By |
|---|---|
| Least Capability | Sydney (violation), Code Review Bot (success) |
| Verification Layers | Near-Miss (saved by verification) |
| Trust Decay | Drift (gradual failure) |
| Adversarial Oversight | Jury Systems, Oversight Dilemma |
| Bounded Exposure | Nuclear Launch |
Next Steps
Section titled “Next Steps”- Apply these lessons → Design Patterns
- Understand the theory → Delegation Risk
- See other domains → Cross-Domain Methods