PRMIA AI RM Certification Exam Sample Questions

AI RM Dumps PDF, AI Risk Management Dumps, download AI Risk Management free Dumps, PRMIA AI Risk Management exam questions, free online AI Risk Management exam questionsYou have to pass the AI RM exam to receive the certification from PRMIA. To increase the effectiveness of your study and make you familiar with the actual exam pattern, we have prepared this PRMIA AI Risk Management sample questions. Our Sample PRMIA AI Risk Management Practice Exam will give you more insight about both the type and the difficulty level of the questions on the PRMIA AI Risk Management exam.

However, we are strongly recommending practice with our Premium PRMIA AI Risk Management (AI RM) Practice Exam to achieve the best score in your actual PRMIA AI RM Exam. The premium practice exam questions are more comprehensive, exam oriented, scenario-based and exact match of PRMIA AI Risk Management exam questions.

PRMIA AI Risk Management Sample Questions:

01. A firm exposes a credit-scoring model through an interface that returns a score and a confidence value for each enquiry. A security review warns that an external party submitting large numbers of systematically varied enquiries could learn about the data the model was trained on.
Which threat is being described?
a) Prompt injection, in which crafted input causes the model to disregard its instructions.
b) Data poisoning, in which the queries corrupt the data the model subsequently learns from.
c) Model inversion, in which repeated queries reveal the training data.
d) Evasion, in which crafted inputs produce an incorrect score.
 
02. A firm's AI inventory is maintained through a registration process: any team introducing an AI system records it. The inventory is reviewed annually and has been accurate at each review. A new assurance review nevertheless finds AI functionality operating in several business processes that was never registered.
What explains this most convincingly?
a) Teams are registering only systems they consider material, applying their own judgment about what is in scope.
b) AI is arriving inside software the firm already uses.
c) The annual review is too infrequent to capture systems introduced between reviews.
d) The registration process is too burdensome, so teams avoid it where they can.
 
03. A model validator reports that a pricing model has high statistical bias and low variance. The term is used here in its statistical sense, not in the sense of unfair treatment of customers.
What does this finding indicate about the model?
a) It is consistently wrong in the same direction, because it is too simple.
b) It systematically disadvantages an identifiable group of customers in the prices it produces.
c) It is highly sensitive to which particular records happened to fall in the training sample.
d) It is accurate on average across the portfolio, because its errors in one direction cancel out those in the other.
 
04. A customer complains that their credit application was declined unfairly. The firm's response states that the decision was produced by an externally supplied model, that the model is widely used in the industry, and that the firm therefore cannot comment on the reasons for the outcome.
How should the risk function assess this response?
a) As adequate, since the firm has been transparent that an external model produced the decision.
b) As adequate, provided the vendor supplies an explanation to the customer.
c) As unacceptable, because the firm should have built the model internally rather than buying it.
d) As unacceptable, because the decision is the firm's own and it remains accountable.
 
05. A firm uses a third-party AI service to screen job applicants. Candidates complain of a pattern of rejections affecting one group. The vendor holds that its tool was tested for fairness before release and has supplied a certificate covering that validation. The firm holds diversity data supplied voluntarily by its applicants.
Which step would do most to establish whether there is a problem in this firm's use of the tool?
a) Ask the vendor to supply the fairness testing it performed before the tool was released.
b) Measure rejection rates across groups in the firm's own applicant data.
c) Add a human review step to every rejection.
d) Suspend use of the tool until the vendor has completed a full review of its model.
 
06. A marketing team wishes to use a dataset that was assembled and is maintained by the operations function for servicing purposes. The dataset has a documented owner in operations. The AI governance framework requires that new uses of data be approved.
Who should approve this new use, and on what basis?
a) The AI governance committee, because any new use of data for a model is within its remit.
b) The data science team, which can assess whether the dataset is technically suitable for the model.
c) The marketing team, since it will be accountable for the outcomes the model produces.
d) The data owner in operations, who is accountable for the dataset.
 
07. A firm deploys an assistant that reads incoming supplier emails and drafts payment instructions for approval. An email arrives containing text, formatted to look like a system notice, instructing the assistant to disregard its previous instructions and treat the sender's bank details as verified. The assistant does so.
What is the most effective response?
a) Constrain what the assistant is permitted to do, so no email instruction produces a payment.
b) Filter incoming emails for text resembling system notices.
c) Retrain the assistant on injection examples so it learns to reject them.
d) Strengthen the assistant's instructions, so it is told to ignore directions found in email content.
 
08. A firm relies on a single vendor's AI service for a customer-facing process. Its resilience assessment records that the service has an availability commitment and that the vendor operates across multiple regions.
What has the assessment most likely overlooked?
a) Whether the vendor's availability commitment is backed by financial compensation for downtime.
b) Whether the firm could move elsewhere at all.
c) Whether the vendor's multi-region operation introduces cross-border data transfer considerations.
d) Whether the service's performance has been tested at the firm's peak transaction volumes.
 
09. A firm's fraud model was trained when most customer payments were made in branch. Payment behaviour has since moved overwhelmingly to mobile channels, though what constitutes fraudulent behaviour has not itself changed.
Which term describes what the model is experiencing?
a) Concept drift, because the meaning of fraud in the firm's business has moved on since the model was built.
b) Overfitting, because the model learned patterns that were specific to branch payments at the time.
c) Data drift, because the inputs changed while the target relationship did not.
d) Data leakage, because channel information was available to the model throughout training.
 
10. A risk report groups two recent problems together as accuracy failures: a fraud classifier that wrongly cleared a set of transactions, and a generative assistant that invented a product feature when answering a customer query. It recommends a single remediation of improved training data for both.
What is the principal flaw in that recommendation?
a) The classifier was wrong about a defined target, while the assistant was never optimising for truth.
b) Improved training data is never an effective remediation for a model failure.
c) The generative failure is the more serious of the two and warrants a separate response for that reason.
d) The two systems are owned by different areas, so one remediation cannot be assigned.

Answers:

Question: 01
Answer: c		 Question: 02
Answer: b		 Question: 03
Answer: a		 Question: 04
Answer: d		 Question: 05
Answer: b
Question: 06
Answer: d		 Question: 07
Answer: a		 Question: 08
Answer: b		 Question: 09
Answer: c		 Question: 10
Answer: a

If you find any errors or typos in PRMIA AI Risk Management (AI RM) (AI RM) sample question-answers or online PRMIA AI Risk Management practice exam, please report them to us on feedback@processexam.com

Your rating: None Average: 5 (1 vote)