Understand
Field Guide
“What do I need to understand?”
Twelve mental models for responsible AI in real institutions, in three arcs: what a deployed system actually is, the specific ways its safety breaks under pressure, and what governing it adaptively looks like.
These pages are conceptual framing — ways of seeing, drawn from the PAN framework's published vocabulary. Documented events live in the Domain Atlas, with citations.
Arc 1
Seeing the system
The parts, the flows, and the memory: mental models for what a deployed AI system actually is.
Sociotechnical systems
ConceptAI safety is a property of the whole working system — the tool, the people, the records, and the rules — not of the model alone.
The system map: nodes, links, and actors
ConceptName the parts and their connections before governing them: the system's components (nodes), the links between them, and the people and bodies who can change them (actors).
Error propagation loops
ConceptErrors do not stay where they are made: they move from models to people to records and back, along a small set of nameable pathways.
Records as living memory
ConceptCase files, databases, and caches are not passive storage: they are an active reservoir that can keep yesterday's errors infecting tomorrow's decisions.
Arc 2
Why safety breaks under pressure
The specific shapes failure takes when evaluated systems meet institutional reality.
The robustness gap
ConceptThe distance between how safe a system looked when it was evaluated and how it actually behaves in a messy, pressured deployment.
Emergent misalignment
ConceptAlignment that held at one capability level, in one context, can quietly stop holding as the system, its users, or its environment change.
AI iatrogenics
ConceptHarm caused by the intervention itself: safeguards and fixes that create new failure modes somewhere else in the system.
Performance ceilings
ConceptSome governance dials cannot be turned past the state of the art — and honest governance plans around the ceiling, not the brochure.
Deskilling and automation bias
ConceptThe human side of the loop degrades quietly: deference drifts up, skills drift down, and the safeguard called 'human oversight' hollows out.
Sycophancy: the people–model feedback loop
ConceptModels that agree harder under pushback, and users who prompt for the answer they hold, close a loop that manufactures false confirmation.
Arc 3
Governing adaptively
What durable governance looks like in a system that will not hold still.
Self-limiting vs self-sustaining error
ConceptThe most important question about a deployment is not how often it errs, but whether its errors die out on their own — or feed themselves.
Governing adaptively
ConceptComplex adaptive systems outrun static rules: durable governance watches, responds, holds authority ready, and rehearses before it prescribes.