Getting Started

This is the orientation page for the whole site. It states the problem, explains why AI systems raise it acutely, sketches the approach, and points you to the right path for your goals and time.

The Problem

Every delegation involves risk. When you delegate a task—to an employee, a contractor, a software system, or an AI agent—you’re accepting potential downside in exchange for capability you don’t have or can’t apply yourself.

This is straightforward but worth stating precisely: Delegation Risk is the expected cost of delegating. For each way things could go wrong (we’ll call these harm modes), multiply the probability by the damage and sum across all harm modes. See the canonical formula in Core Concepts.

A research assistant that might leak proprietary data (P=0.001, damage=$50,000) contributes $50 to your Delegation Risk. An AI code deployer that might deploy malicious code (P=0.0001, damage=$1,000,000) contributes $100. You’re implicitly accepting these costs when you delegate.

This framing has two virtues:

It’s quantifiable. You can compare delegates, track changes over time, and make principled trade-offs. “Is this AI system safer?” becomes a question with a numerical answer—at least in principle, and sometimes in practice.
It separates the problem from the solution. You might reduce Delegation Risk by selecting more trustworthy delegates, by adding verification layers, or by limiting what delegates can do. The metric doesn’t assume a particular approach.

Of course, the hard part is estimating the probabilities and damages. We’ll address this, but it’s worth acknowledging upfront: Delegation Risk is only as good as your estimates. The framework provides structure for reasoning about risk; it doesn’t eliminate uncertainty.

Why AI Systems?

AI systems may present delegation challenges at unusual scale. Several factors suggest this:

Capabilities are expanding rapidly, and verification may not be keeping pace. A year ago, AI couldn’t reliably write working code; now it often can. Our tools for verifying AI behavior have improved, but it’s not obvious they’ve improved proportionally.

Agent-to-agent delegation creates complex risk networks. When AI agents delegate to other AI agents, risk relationships become networks rather than chains. If your AI assistant uses a plugin that calls another API that invokes another model—how much Delegation Risk are you accepting? Most current systems don’t have principled answers.

Autonomy may increase while oversight decreases. There are economic pressures toward more autonomous AI (Gwern, 2016). To the extent autonomy increases without proportional accountability, structural guarantees become more important.

We may be building systems we don’t fully understand. Uncharacterized capabilities mean unenumerated harm modes—and the Delegation Risk calculation requires listing failure modes, so uncertainty about capabilities translates directly to uncertainty about risk.

None of these claims are certain. But if even some hold, infrastructure for managing Delegation Risk seems valuable. And even if AI turns out easier to manage than feared, the framework applies to delegation generally—the AI application is primary but not exclusive.

The Approach, in One Paragraph

The framework proposes structural constraints: design systems where dangerous behavior is difficult or impossible regardless of intentions, so that safety is a property of the architecture rather than of any component’s behavior. This is how nuclear plants budget failure probabilities, how secure systems limit blast radius, and the logic behind constitutional separation of powers. Core Concepts develops the idea in full—decomposition, the “Least X” principles, and what the framework does and doesn’t provide.

Two pointers worth following early:

Scope: this is about containment, not alignment—it complements alignment research, interpretability, and AI Control rather than replacing them. For an explicit list of what the framework does not provide, see What This Framework Is NOT.
Generality: the same mathematics applies to organizational trust, software permissions, supply chains, and government delegation. We use these domains throughout because human-delegation intuitions are often sharper than AI ones, and decades of practice provide tested patterns.

Choose Your Path

The Core Path is the featured route—the framework’s strongest material in order. Everything else branches by goal and time.

Path	For you if you want to…	Time
⭐ The Core Path	read the best of the site in order: entanglements, risk decomposition, channel integrity, and four worked examples (~13 stops). If you read one route, read this.	4–6 hrs
Five-Minute Intro	get the essential concepts on one page — start here	5 min
+ Core Concepts and FAQ	evaluate whether this framework is useful	30–60 min
Core Concepts → Delegation Risk → Quick Start	apply the framework to a real system	2–3 hrs
For Engineers	implement it: patterns, checklists, debugging	15 min

Reference pages—Quick Reference, Glossary, Common Mistakes, Examples Catalog—are linked from the sidebar when you need them.