Home / AWS AI Practitioner

AWS · AIF-C01

AWS AI Practitioner Practice Exam (AIF-C01)

Free AWS AI Practitioner practice exam questions from all 5 official domains, with every answer option explained.

Start practicing free Take a mock examSee plans

AWS Certified AI Practitioner (AIF-C01)

Questions
65
Time
90 min
To pass
700 / 1000

Scored on a 100-1000 scale.

Exam Price
$100

What is on the AIF-C01 exam?

Open a domain for its objectives and their practice questions.

Try free AIF-C01 practice questions

Pick an answer to see why each option is right or wrong. A preview of 4 of the 10 free questions, which cover every domain.

Domain 1: Fundamentals of AI and ML

1.2 Identify practical use cases for AIeasy

A global support team receives written chat messages in about a dozen languages. The agents read only English and need to read every message in English. Which AWS service meets this need?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Amazon Polly speaks text aloud. It would read the messages in their original languages, which the agents still could not understand.

Amazon Comprehend is tempting because it works with multilingual text and can identify the language of each message. It extracts insights such as language, entities and sentiment, but it does not render the message in English.

Amazon Translate is a text translation service that converts text between languages on demand, so each message can be shown to agents in English.

Amazon Transcribe converts spoken audio into text. The chat messages are already written, so there is no audio to transcribe.

1.1 Explain basic AI concepts and terminologieseasy

An internal tool calls a model only a few times a day, at unpredictable moments. The team wants to avoid managing instances or scaling policies and to pay nothing for idle time. Which SageMaker AI inference option fits?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Running the model on your own server is the opposite of the requirement: the team would provision, scale and pay for the machine whether or not it is called.

Real-time inference gives low latency, which makes it tempting for any live call. It runs a persistent endpoint on an instance type you choose and is meant for sustained traffic, so this team would manage instances and pay while the tool sits idle.

Batch transform processes a large dataset that is available upfront, offline. These calls arrive one at a time at unpredictable moments, not as a dataset prepared in advance.

Serverless inference is built for intermittent or unpredictable traffic: SageMaker AI manages the infrastructure, so there are no instances or scaling policies to manage, and you pay only for use, not idle time.

Domain 2: Fundamentals of GenAI

2.2 Understand the capabilities and limitations of GenAI for solving business problemseasy

A legal team needs a model that can take in an entire lengthy merger agreement in one request and answer questions about how its sections relate. Which model characteristic matters most?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Image generation is a different modality; the task is reading and reasoning over text.

The context window is how much text the model can consider at once; a larger one lets it process the whole agreement in a single run.

Parameter count indicates model complexity and capacity to learn patterns, which makes it tempting, but it does not set how much text fits into one request.

Price matters for budget, but a cheap model with a small window still cannot take in the whole agreement.

2.1 Explain the basic concepts of generative AI (GenAI)easy

An insurer wants to train an image model that classifies hail damage on roofs, but collecting and labeling enough real roof photos is slow and costly. Which generative AI use case helps most with this?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Code generation is a genuine productivity use case for developers. Faster preprocessing still leaves the same small set of photos, so the data shortage remains.

Generative models can create synthetic training data, including labeled examples for supervised learning, when real data is expensive or hard to collect. Extra labeled images address the insurer's actual shortage.

Encryption protects stored data but is not a generative AI use case, and it produces no additional training examples.

Summarization is a real generative AI use case for knowledge search, which makes it sound useful. It condenses existing notes, though, and adds no new labeled images for the classifier to learn from.

The other 6 free questions, by domain:

Domain 3: Applications of Foundation Models · 2 questions
3.2 Choose effective prompt engineering techniqueseasy

A designer uses Amazon Nova to generate images of a mountain bike on a forest trail. The images keep showing other riders in the background, which the designer does not want. Which prompt technique addresses this?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Few-shot examples are input-output pairs that calibrate text output. They are not the mechanism for keeping an object out of a generated image.

Chain-of-thought breaks a reasoning problem into intermediate steps. It is for multi-step reasoning tasks, not for excluding objects from an image.

Negative prompts, set through the negativeText parameter, exclude objects or styles that would otherwise tend to appear as a result of the main prompt, such as background riders on a trail.

More randomness changes the results unpredictably. It does not tell the model to leave out a specific element.

3.1 Describe design considerations for applications that use foundation models (FMs)easy

A developer wants the model to stop generating as soon as it writes the marker "###" at the end of each record. Which inference setting provides this behavior?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Temperature makes output more or less deterministic. It has no notion of a marker at which to stop.

Penalties discourage repeated or frequent tokens in a response. They do not end the output when a particular marker appears.

Top P controls how much of the probability distribution the model samples from for each token. It changes word choice, not where generation ends.

A stop sequence is a string that tells the model to stop generating further tokens once it produces that string, which is exactly the behavior wanted for the record marker.

Domain 4: Guidelines for Responsible AI · 2 questions
4.1 Explain the development of AI systems that are responsibleeasy

A health insurer's member assistant on Amazon Bedrock must not engage in conversations about specific medical treatment advice. The team can describe that subject in plain language. Which Amazon Bedrock Guardrails safeguard fits this requirement?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Word filters block exact custom words or phrases; a subject described in plain language can be phrased many ways, which a fixed word list cannot cover.

Grounding checks catch responses not supported by a source or irrelevant to the query; a well-grounded answer about treatment would still pass.

Denied topics let you define subjects that are undesirable in your application, and the guardrail blocks them when they appear in user queries or model responses.

Content filters detect predefined harmful categories such as hate or violence; treatment advice is not one of those categories.

4.2 Recognize the importance of transparent and explainable modelseasy

Which description fits the explainable machine learning techniques, such as LIME and SHAP, that AWS groups apart from interpretable ML?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Blocking harmful output is the job of safeguards such as guardrails, not of explanation techniques.

That describes the other category, interpretable ML, such as expressive Boolean formulas, which relies on models whose workings can be checked directly.

These techniques explain a model's decisions after they are made, so they are post hoc explanations of models whose inner workings are hidden.

Comparing generated summaries against human references is a model evaluation task, not a way to explain a model's decisions.

Domain 5: Security, Compliance, and Governance for AI Solutions · 2 questions
5.1 Explain methods to secure AI systemseasy

A manufacturer's analysts lose weeks at the start of each ML project working out which datasets exist and who owns them, because the data sits in separate team systems with no shared inventory. Which practice addresses this?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

GuardDuty watches AWS logs for unexpected or malicious activity. That protects the environment but does not tell analysts what datasets exist or who owns them.

KMS keys protect data at rest, which is good practice, but encrypted data is no easier to find or to trace to an owner.

A data catalog gives a unified view of data assets through business and technical metadata and identifies data owners, which cuts the time spent searching for the right data.

Training settings act on data a team already has. The delay happens before training, while analysts are still searching for the data.

5.2 Recognize governance and compliance regulations for AI systemseasy

A team runs model-serving code on Amazon EC2 instances and in AWS Lambda functions. It wants continuous, automatic scanning of these workloads for software vulnerabilities and unintended network exposure. Which service fits?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Model cards document a model's purpose and risk. They describe the model rather than scanning the compute it runs on.

Inspector is a vulnerability management service that discovers workloads such as EC2 instances and Lambda functions and continually scans them for software vulnerabilities and network exposure.

CloudTrail records actions taken in the account, which is useful after an incident, but it does not look for vulnerable software packages.

Artifact supplies AWS compliance documents. It does not scan any of the team's workloads.

Practice all 10 free questionsGet 150+ questions

AIF-C01 question bank coverage

DomainExam weightFree sampleMaster (full bank)
1. Fundamentals of AI and ML20%239
2. Fundamentals of GenAI24%247
3. Applications of Foundation Models28%255
4. Guidelines for Responsible AI14%227
5. Security, Compliance, and Governance for AI Solutions14%227
Total100%10195

Looking for a AIF-C01 practice exam? Start with the free questions above, then take the timed mock exam to rehearse the AIF-C01 format under exam conditions.

How these AIF-C01 questions are made

The domains, weights and objectives come from AWS's official AIF-C01 exam guide (v1.2). Questions are original and written only from the vendor's official prep material, and each one cites the passage it is based on. Every question is then answered independently, without seeing its key, and every answer option's explanation is checked for accuracy before it is published.