AWS Certified AI Practitioner (AIF-C01)
Practice Exam 2026 | Questions &
Answers with Detailed Rationales | AI
Fundamentals, Generative AI, Machine
Learning Concepts, Responsible AI & AWS
AI Services | Verified Edition
1. A company wants to use a large language model (LLM) to
summarize customer support tickets. The company wants to
minimize the risk of the model generating information that is
not present in the source ticket. Which approach is MOST
appropriate?
A. Increase the model temperature
B. Increase the maximum token limit
C. Use retrieval-augmented generation (RAG) with the ticket
content as context
D. Fine-tune the model using unrelated public documents
Answer: C. Use retrieval-augmented generation (RAG) with
the ticket content as context
Rationale: RAG supplies the model with relevant source
information at inference time, allowing the generated
summary to be grounded in the retrieved ticket content.
Increasing temperature generally increases variability and
,does not reduce hallucinations. A larger token limit affects
output capacity rather than factual grounding, while
unrelated fine-tuning data does not ensure that responses
remain faithful to the ticket.
2. An organization is evaluating an AI application that
generates product descriptions. The organization wants to
measure how closely generated descriptions match reference
descriptions while considering word overlap. Which metric is
MOST appropriate?
A. Recall
B. BLEU
C. Mean squared error
D. ROC-AUC
Answer: B. BLEU
Rationale: BLEU is commonly used to evaluate machine-
generated text by comparing generated output with one or
more reference texts using n-gram precision and a brevity
penalty. Recall and ROC-AUC are generally associated with
classification tasks, while mean squared error is commonly
used for regression.
3. A financial institution uses an ML model to approve or
reject loan applications. Historical training data contains
significantly more approved applications than rejected
applications. Which problem could this imbalance MOST
directly create?
,A. Increased model interpretability
B. Improved minority-class recall
C. Poor performance on the minority class
D. Elimination of training bias
Answer: C. Poor performance on the minority class
Rationale: Class imbalance can cause a model to favor the
majority class, potentially producing strong overall accuracy
while performing poorly on the minority class. This is
especially important in high-impact applications because
aggregate accuracy can hide poor recall or precision for the
less frequent class.
4. A company wants to allow users to interact with a
generative AI application while preventing users from
submitting prompts containing confidential employee
information. Which control is MOST directly relevant?
A. Prompt filtering and input validation
B. Increasing model temperature
C. Increasing inference batch size
D. Reducing the model context window
Answer: A. Prompt filtering and input validation
Rationale: Input validation and filtering can detect or block
sensitive information before it reaches the model.
Temperature affects generation randomness, batch size
affects processing efficiency, and context-window size controls
how much information can be processed rather than directly
preventing confidential information from being submitted.
, 5. A company wants an AI system to answer questions about
its internal HR policies. The policies change frequently, and
the company does not want to retrain the foundation model
whenever a policy changes. Which architecture is MOST
appropriate?
A. Train a new foundation model for every policy update
B. Use RAG to retrieve the latest HR policy information
C. Increase model temperature after every policy update
D. Remove the HR documents from the application
Answer: B. Use RAG to retrieve the latest HR policy
information
Rationale: RAG allows an application to retrieve current
information from an external knowledge source and provide
that information to the model during inference. This is
particularly useful when information changes frequently
because the underlying model does not need to be retrained
for every document update.
6. A generative AI application produces different answers to
the same prompt because the application uses a high
temperature value. What does temperature primarily
control?
A. The amount of training data
B. The randomness of generated output
C. The number of model parameters
D. The size of the model's training dataset
Practice Exam 2026 | Questions &
Answers with Detailed Rationales | AI
Fundamentals, Generative AI, Machine
Learning Concepts, Responsible AI & AWS
AI Services | Verified Edition
1. A company wants to use a large language model (LLM) to
summarize customer support tickets. The company wants to
minimize the risk of the model generating information that is
not present in the source ticket. Which approach is MOST
appropriate?
A. Increase the model temperature
B. Increase the maximum token limit
C. Use retrieval-augmented generation (RAG) with the ticket
content as context
D. Fine-tune the model using unrelated public documents
Answer: C. Use retrieval-augmented generation (RAG) with
the ticket content as context
Rationale: RAG supplies the model with relevant source
information at inference time, allowing the generated
summary to be grounded in the retrieved ticket content.
Increasing temperature generally increases variability and
,does not reduce hallucinations. A larger token limit affects
output capacity rather than factual grounding, while
unrelated fine-tuning data does not ensure that responses
remain faithful to the ticket.
2. An organization is evaluating an AI application that
generates product descriptions. The organization wants to
measure how closely generated descriptions match reference
descriptions while considering word overlap. Which metric is
MOST appropriate?
A. Recall
B. BLEU
C. Mean squared error
D. ROC-AUC
Answer: B. BLEU
Rationale: BLEU is commonly used to evaluate machine-
generated text by comparing generated output with one or
more reference texts using n-gram precision and a brevity
penalty. Recall and ROC-AUC are generally associated with
classification tasks, while mean squared error is commonly
used for regression.
3. A financial institution uses an ML model to approve or
reject loan applications. Historical training data contains
significantly more approved applications than rejected
applications. Which problem could this imbalance MOST
directly create?
,A. Increased model interpretability
B. Improved minority-class recall
C. Poor performance on the minority class
D. Elimination of training bias
Answer: C. Poor performance on the minority class
Rationale: Class imbalance can cause a model to favor the
majority class, potentially producing strong overall accuracy
while performing poorly on the minority class. This is
especially important in high-impact applications because
aggregate accuracy can hide poor recall or precision for the
less frequent class.
4. A company wants to allow users to interact with a
generative AI application while preventing users from
submitting prompts containing confidential employee
information. Which control is MOST directly relevant?
A. Prompt filtering and input validation
B. Increasing model temperature
C. Increasing inference batch size
D. Reducing the model context window
Answer: A. Prompt filtering and input validation
Rationale: Input validation and filtering can detect or block
sensitive information before it reaches the model.
Temperature affects generation randomness, batch size
affects processing efficiency, and context-window size controls
how much information can be processed rather than directly
preventing confidential information from being submitted.
, 5. A company wants an AI system to answer questions about
its internal HR policies. The policies change frequently, and
the company does not want to retrain the foundation model
whenever a policy changes. Which architecture is MOST
appropriate?
A. Train a new foundation model for every policy update
B. Use RAG to retrieve the latest HR policy information
C. Increase model temperature after every policy update
D. Remove the HR documents from the application
Answer: B. Use RAG to retrieve the latest HR policy
information
Rationale: RAG allows an application to retrieve current
information from an external knowledge source and provide
that information to the model during inference. This is
particularly useful when information changes frequently
because the underlying model does not need to be retrained
for every document update.
6. A generative AI application produces different answers to
the same prompt because the application uses a high
temperature value. What does temperature primarily
control?
A. The amount of training data
B. The randomness of generated output
C. The number of model parameters
D. The size of the model's training dataset
