liveBook · Manning

Risk ladder

As explained in AI Governance

The risk ladder is the book’s framework for showing how risks accumulate as an AI application gains capabilities.

A risk ladder tracks how exposure grows as an AI application gains capabilities. Use it to check whether controls still match the system’s environment, inputs, model, data access, ability to act, and autonomy. section 4.2.1

Table 4.1 How each additional ClaimAssist feature increases the security risk
Version
Environment
Input
Model
Data
Ability to Act
Autonomy
V1: FAQ Bot
Yes
Yes
No
No
No
No
V2: FAQ Bot running on an unvetted model
Yes
Yes
Yes
No
No
No
V3: Personalized Assistant
Yes
Yes
Yes
Yes
No
No
V4: Fine-tuned Assistant
Yes
Yes
Yes
Yes
No
No
V5: Action-Enabled Assistant
Yes
Yes
Yes
Yes
Yes
No
V6: Agentic Assistant
Yes
Yes
Yes
Yes
Yes
Yes

Compare six ClaimAssist versions, from an FAQ bot to an agentic assistant, to see which risk factors appear as capabilities are added. section 4.2.1

Agent trust can widen exposure

Hidden instructions in a claim can pass to a scheduling agent. If it trusts the requesting agent, it may retrieve the adjuster’s full calendar, including other claimants’ visit addresses. Controls suitable for a FAQ bot may not cover tool calls or state changes. section 4.2.1

How AI Governance covers it

How deep it goes
Compares six capability stages and identifies control needs, while detailed factor analysis is deferred.
Examples
Uses a progression from FAQ bot to agentic assistant and a hidden-instruction attack involving calendar data.
What you can do after
Evaluate whether controls match exposure as an assistant gains data access, action capability, and autonomy.
How it connects
Connects input validation and logging with tool calls, state changes, and trust relationships between agents.

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.