Open Assessment · RAIL
Responsible AI Leader practice test
20 free questions, mostly leadership scenarios, across the seven RAIL domains, with the correct answer and the reasoning explained. Open all year, as many times as you like. It grants no certificate or badge: it is preparation for the official RAIL exam (50 questions in 150 minutes, in English, Portuguese or Spanish).
Start the assessmentFree
No account, no card, no limit on attempts.
Not the certification exam
Separate questions, never reused in the exam.
Nothing on your record
Your answers are not kept. Only the result by domain is saved, linked to you only if you are signed in, and it produces no credential.
Sample questions
Two questions from this practice test, in the same style as the certification exam. Answer before opening the explanation.
Question 1
The charter of a payments company's AI committee says it "reviews and advises on" AI use cases. In its first year, two business units launched high-risk systems against its recommendations, and nobody had to justify the decision. Which change to the charter most directly addresses this?
- Make the chief executive the committee chair, so that its recommendations carry more weight with business heads
- Increase the meeting frequency, so that recommendations reach business units well before their launch dates
- Give the committee defined decision rights over high-risk use cases, with a documented escalation route for disputes
- Require business units to attach the committee's recommendation to the documentation of each launch
Show answer and explanation
Correct answer: C. Give the committee defined decision rights over high-risk use cases, with a documented escalation route for disputes
The gap is authority: an advisory mandate lets business units ignore the committee without consequence. A charter should state decision rights, for example approve, approve with conditions or reject for defined risk tiers, and how disagreements escalate, such as to the executive committee, so that overriding the committee becomes a visible and accountable decision. A senior chair adds influence, but influence is not a decision right and the mandate stays advisory. More meetings deliver advice sooner that can still be ignored. Attaching the recommendation records the disagreement but requires no one to resolve it.
Question 2
The AI policy requires a bias evaluation before any customer-facing model goes live. A product team asks to launch a pricing model without it, because the evaluation dataset will only be ready later. Which way of handling the request reflects good governance practice?
- A documented, time-limited exception approved at an authority level that matches the risk, with compensating controls and a named owner
- A refusal in every case, since allowing exceptions to a mandatory requirement weakens the authority of the policy
- Approval by the governance office alone, provided the team commits in writing to run the evaluation afterwards
- Approval if the team lead accepts the risk by email, since the team knows the model and its customers best
Show answer and explanation
Correct answer: A. A documented, time-limited exception approved at an authority level that matches the risk, with compensating controls and a named owner
A policy needs an exception route, or teams work around it silently. A sound exception is documented, has an expiry date, is approved by someone whose authority is proportionate to the risk, comes with compensating controls such as a limited rollout or closer monitoring, and is entered in a register so exceptions can be tracked and reported. An absolute refusal sounds firm but tends to push teams to bypass the process. The governance office approving alone takes risk ownership away from the business, and a written promise is not a compensating control. A team lead's email is acceptance at too low a level and leaves no trace in the register.