Skip to content
AI Security

Interview question · Agent, Tool & MCP Security

How would you measure an agent's excessive-agency risk quantitatively, so you can compare designs and track it over time?

Separate reachable authority, forbidden proposals, gate decisions and confirmed effects. Compare designs with matched tasks, explicit denominators and uncertainty; tool counts and zero observed failures are not universal risk scores.

hardNewUpdated Nov 2026

Separate reachable authority, forbidden proposals, gate decisions and confirmed effects. Compare designs with matched tasks, explicit denominators and uncertainty; tool counts and zero observed failures are not universal risk scores.

Make your first attempt

What would you say before reading the answer?

  1. 01 State the assumptions your answer depends on.
  2. 02 Explain the mechanism and choose a control.
  3. 03 Name a condition that would change your decision.

The worked answer includes an approach, scoring criteria and follow-up questions. Compare the reasoning with your own.

10 answers per topic instead of 5, plus saved progress and bookmarks · costs nothingor unlock all 529 remaining answers · $25

Every worked answer in all 17 topics, every lesson in every course, the full concept curriculum, the complete practice-test bank, and the downloadable security field guides.

6 months · One payment · No auto-renewal

What if · the follow-ups in this answer

The answer works through 4 follow-ups like an interviewer would

In a synthetic 200-run evaluation, 40 runs produce a forbidden proposal. Policy blocks 30; reviewers reject six of the remaining ten. Three dispatched effects are confirmed and one is unresolved.

  1. What is the observed reviewer rejection fraction?
  2. Can the report claim that 197 of 200 runs prevented the forbidden effect?
  3. Does zero failure in 300 adaptive attempts prove a failure rate below 1%?
  4. What makes a design comparison useful?

Each rung has a nudge, a model answer, the principle that carries over and the weak answer to avoid. Premium opens every answer and ladder.

UP NEXT ON YOUR JOURNEY
Next in this trackWhat is scope drift in an autonomous agent, and how would you detect it from the agent's traces?Next in this trackBeyond a single injected prompt, how do attackers corrupt AI agents through memory, tools, retrieved knowledge and other agents?Next in this trackHow do AGENTS.md, rules files and editor or MCP config in a repository become an attack path against coding agents?
DISCUSSION · 0

No comments yet. Share a correction or a different way to answer it.