ITIL 5: Responsible AI Practice practice questions
27 questions on Responsible AI Practice, one of the 11 areas assessed in ITIL 5 Foundation. Untimed, with an explanation after every answer.
ITIL 5 describes itself as designed for AI, and this topic tests how to use AI in services responsibly: a named human accountable for a model’s outcomes, a route to contest decisions, human in the loop, transparency, governance of training data, bias through proxy variables, model drift, and a register of the AI systems in use.
Accountability is never outsourced along with a third-party model. Disclosing that AI was used is not enough when a decision materially affects someone. Removing a protected attribute does not remove bias, because proxies reproduce it. Deflection is not resolution. Automating a poor process produces bad outcomes faster, and human oversight should scale with the consequence of an error.
The ITIL 5 exam is also offered in French, among other languages; these practice questions are in English. Which languages? →
An organisation deploys an AI model that automatically approves or rejects low-value expense claims. Which control most directly supports accountability for its decisions?
Ready for exam conditions? 40 questions, 60 minutes, 65% to pass.
Mock examA worked example
A question from the bank with its answer and the reasoning behind it. Every practice question comes with an explanation like this one.
An AI assistant embedded in a service desk begins giving confident but incorrect answers about a recently changed policy. Which control would have most directly limited the harm?
- AIncreasing the model's response speed
- BGrounding responses in a maintained knowledge source and showing the source with each answer(correct answer)
- CMaking the assistant's tone more formal
- DRestricting the assistant to a smaller group of users
Why
The failure is unsourced generation drifting from current policy. Grounding answers in a maintained authoritative source and surfacing that source with each answer keeps content current and lets the user verify it. Speed (A) and tone (C) are irrelevant to correctness, and restricting the audience (D) reduces exposure without addressing the defect.