Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 63

Question 620

A reviewer of a telecommunications AI use case needs to prevent compromised retrieval material from manipulating the answer. Which term identifies the situation?

  1. Prompt injection recognition
  2. Amazon Bedrock Prompt Management
  3. Prompt versioning
  4. Prompt poisoning

Explanation

Correct Answer

D:

This is the choice aligned with the telecommunications requirement to prevent compromised retrieval material from manipulating the answer: poisoned source content is deliberately planted to influence downstream model output. A realistic pilot must still check the qualification that editing a trusted instruction cannot repair every compromised external source, especially where that could affect the acceptance criterion.

Incorrect Answers

A:

For the telecommunications objective to prevent compromised retrieval material from manipulating the answer, Prompt injection recognition addresses a different task—detect instructions hidden in retrieved web pages that attempt to override the user task. Its operating mechanism differs from that of Prompt poisoning: prompt injection abuses lower-trust content to redirect a model's behavior; this does not satisfy the requested outcome.

B:

For the telecommunications objective to prevent compromised retrieval material from manipulating the answer, Amazon Bedrock Prompt Management addresses a different task—centrally manage and version reusable FM prompts. Its operating mechanism differs from that of Prompt poisoning: Bedrock Prompt Management supports prompt variants and managed lifecycle for supported workflows; this does not satisfy the requested outcome.

C:

For the telecommunications objective to prevent compromised retrieval material from manipulating the answer, Prompt versioning addresses a different task—reproduce and roll back a prompt that behaved better last week. Its operating mechanism differs from that of Prompt poisoning: versioned prompts preserve tested text and configuration across deployments; this does not satisfy the requested outcome.

Question 621

The media publishing acceptance review includes Negative prompting for a plan to tell an image or text generator what elements to avoid. Identify THREE accurate statements. (Choose THREE.)

  1. Negative prompting eliminates its documented limitation, so an additional risk check is unnecessary.
  2. This approach directly addresses a requirement to tell an image or text generator what elements to avoid.
  3. Its core purpose is to centrally manage and version reusable FM prompts instead of meeting the stated requirement.
  4. Negative guidance identifies unwanted styles, topics or content traits.
  5. Its operating process follows this proposed account: A single example demonstrates the target response pattern inside the prompt.
  6. A prohibition in prompt text is not a reliable replacement for output validation.

Explanation

Correct Answers

B:

The claim matches the intended use of Negative prompting, namely to tell an image or text generator what elements to avoid. Negative guidance identifies unwanted styles, topics or content traits; therefore this is a real fit for the media publishing task, provided the team accounts for the fact that a prohibition in prompt text is not a reliable replacement for output validation.

D:

This accurately identifies the operation behind Negative prompting: negative guidance identifies unwanted styles, topics or content traits. It offers the mechanism needed to tell an image or text generator what elements to avoid in the media publishing example, with the boundary that a prohibition in prompt text is not a reliable replacement for output validation.

F:

This is a valid caution for Negative prompting: a prohibition in prompt text is not a reliable replacement for output validation. In a media publishing setting, the concern follows even when negative guidance identifies unwanted styles, topics or content traits; it must be checked against the particular requirement to tell an image or text generator what elements to avoid.

Incorrect Answers

A:

This assertion incorrectly waives the limitation of Negative prompting in the media publishing context. The relevant boundary is that a prohibition in prompt text is not a reliable replacement for output validation; selecting a technology to tell an image or text generator what elements to avoid does not remove that underlying requirement.

C:

This substitutes an unrelated objective for the one actually served by Negative prompting. Its purpose in media publishing is to tell an image or text generator what elements to avoid, and the relevant mechanism is that negative guidance identifies unwanted styles, topics or content traits.

E:

The option is not consistent with the behavior of Negative prompting in a media publishing initiative to tell an image or text generator what elements to avoid. The applicable mechanism is that negative guidance identifies unwanted styles, topics or content traits; evaluate the boundary that a prohibition in prompt text is not a reliable replacement for output validation.

Question 622

The subscription services committee must distinguish true from false claims about Jailbreak attempt recognition and the need to recognize requests framed to bypass an assistant's content and safety constraints. Which TWO statements are correct? (Choose TWO.)

  1. The main purpose of Jailbreak attempt recognition is to detect instructions hidden in retrieved web pages that attempt to override the user task instead of to recognize requests framed to bypass an assistant's content and safety constraints.
  2. Jailbreak attempt recognition operates according to this purported mechanism: A system-level instruction establishes high-priority guidance for the model conversation.
  3. A creative role-play wrapper does not increase a user's actual authorization.
  4. Jailbreaks try to persuade a model to ignore governing behavior constraints.
  5. The described method removes the need to consider the stated limitation because grounding instructions establish evidence boundaries and a fallback when sources are missing.

Explanation

Correct Answers

C:

This is a valid caution for Jailbreak attempt recognition: a creative role-play wrapper does not increase a user's actual authorization. In a subscription services setting, the concern follows even when jailbreaks try to persuade a model to ignore governing behavior constraints; it must be checked against the particular requirement to recognize requests framed to bypass an assistant's content and safety constraints.

D:

This accurately identifies the operation behind Jailbreak attempt recognition: jailbreaks try to persuade a model to ignore governing behavior constraints. It offers the mechanism needed to recognize requests framed to bypass an assistant's content and safety constraints in the subscription services example, with the boundary that a creative role-play wrapper does not increase a user's actual authorization.

Incorrect Answers

A:

The stated behavior is for Prompt injection recognition, not for Jailbreak attempt recognition in a subscription services requirement to recognize requests framed to bypass an assistant's content and safety constraints. Prompt injection abuses lower-trust content to redirect a model's behavior is a real capability of the alternative, whereas the concept under review works because jailbreaks try to persuade a model to ignore governing behavior constraints.

B:

The stated behavior is for System instructions, not for Jailbreak attempt recognition in a subscription services requirement to recognize requests framed to bypass an assistant's content and safety constraints. A system-level instruction establishes high-priority guidance for the model conversation is a real capability of the alternative, whereas the concept under review works because jailbreaks try to persuade a model to ignore governing behavior constraints.

E:

This assertion incorrectly waives the limitation of Jailbreak attempt recognition in the subscription services context. The relevant boundary is that a creative role-play wrapper does not increase a user's actual authorization; selecting a technology to recognize requests framed to bypass an assistant's content and safety constraints does not remove that underlying requirement.

Question 623

The industrial maintenance acceptance criterion is to request predictable JSON fields for an automated downstream consumer. Which of these choices should be shortlisted first?

  1. Constrained output schema
  2. Task-specific instructions
  3. Grounded answering instructions
  4. Prompt poisoning
  5. Prompt injection recognition

Explanation

Correct Answer

A:

This is the choice aligned with the industrial maintenance requirement to request predictable JSON fields for an automated downstream consumer: a named schema and response validation establish required structure and allowed fields. A realistic pilot must still check the qualification that prompt wording cannot replace parsing, schema validation or error handling, especially where that could affect the acceptance criterion.

Incorrect Answers

B:

For the industrial maintenance objective to request predictable JSON fields for an automated downstream consumer, Task-specific instructions addresses a different task—improve an ambiguous request by stating the task and success criteria. Its operating mechanism differs from that of Constrained output schema: specific instructions describe what to do, for whom and in what output form; this does not satisfy the requested outcome.

C:

For the industrial maintenance objective to request predictable JSON fields for an automated downstream consumer, Grounded answering instructions addresses a different task—direct the assistant to use only retrieved passages for a policy answer. Its operating mechanism differs from that of Constrained output schema: grounding instructions establish evidence boundaries and a fallback when sources are missing; this does not satisfy the requested outcome.

D:

For the industrial maintenance objective to request predictable JSON fields for an automated downstream consumer, Prompt poisoning addresses a different task—prevent compromised retrieval material from manipulating the answer. Its operating mechanism differs from that of Constrained output schema: poisoned source content is deliberately planted to influence downstream model output; this does not satisfy the requested outcome.

E:

For the industrial maintenance objective to request predictable JSON fields for an automated downstream consumer, Prompt injection recognition addresses a different task—detect instructions hidden in retrieved web pages that attempt to override the user task. Its operating mechanism differs from that of Constrained output schema: prompt injection abuses lower-trust content to redirect a model's behavior; this does not satisfy the requested outcome.

Question 624

An industrial maintenance team must show one completed classification example before a new input. Which listed concept or AWS capability directly matches that requirement?

  1. Task-specific instructions
  2. Chain-of-thought-style decomposition
  3. One-shot prompting
  4. Amazon Bedrock Prompt Management

Explanation

Correct Answer

C:

This is the choice aligned with the industrial maintenance requirement to show one completed classification example before a new input: a single example demonstrates the target response pattern inside the prompt. A realistic pilot must still check the qualification that one example may not cover ambiguous edge cases or unusual categories, especially where that could affect the acceptance criterion.

Incorrect Answers

A:

For the industrial maintenance objective to show one completed classification example before a new input, Task-specific instructions addresses a different task—improve an ambiguous request by stating the task and success criteria. Its operating mechanism differs from that of One-shot prompting: specific instructions describe what to do, for whom and in what output form; this does not satisfy the requested outcome.

B:

For the industrial maintenance objective to show one completed classification example before a new input, Chain-of-thought-style decomposition addresses a different task—improve a multi-step reasoning task through intermediate checks or subquestions. Its operating mechanism differs from that of One-shot prompting: breaking a complex task into smaller verifiable steps can improve task execution; this does not satisfy the requested outcome.

D:

For the industrial maintenance objective to show one completed classification example before a new input, Amazon Bedrock Prompt Management addresses a different task—centrally manage and version reusable FM prompts. Its operating mechanism differs from that of One-shot prompting: Bedrock Prompt Management supports prompt variants and managed lifecycle for supported workflows; this does not satisfy the requested outcome.

Question 625

A new insurance operations initiative has one measurable requirement: reuse a tested set of instructions with different customer details. What should the team select?

  1. Structured prompt template
  2. System instructions
  3. Chain-of-thought-style decomposition
  4. Zero-shot prompting

Explanation

Correct Answer

A:

This is the choice aligned with the insurance operations requirement to reuse a tested set of instructions with different customer details: a template separates stable instructions from variables inserted for each request. A realistic pilot must still check the qualification that templates need version control and sanitization of untrusted variable content, especially where that could affect the acceptance criterion.

Incorrect Answers

B:

For the insurance operations objective to reuse a tested set of instructions with different customer details, System instructions addresses a different task—set assistant behavior and constraints independently of a user question. Its operating mechanism differs from that of Structured prompt template: a system-level instruction establishes high-priority guidance for the model conversation; this does not satisfy the requested outcome.

C:

For the insurance operations objective to reuse a tested set of instructions with different customer details, Chain-of-thought-style decomposition addresses a different task—improve a multi-step reasoning task through intermediate checks or subquestions. Its operating mechanism differs from that of Structured prompt template: breaking a complex task into smaller verifiable steps can improve task execution; this does not satisfy the requested outcome.

D:

For the insurance operations objective to reuse a tested set of instructions with different customer details, Zero-shot prompting addresses a different task—ask a model to complete a standard task without sample answers. Its operating mechanism differs from that of Structured prompt template: zero-shot prompting relies on the model's existing general capabilities; this does not satisfy the requested outcome.

Question 626

In industrial maintenance, the planned workflow needs to centrally manage and version reusable FM prompts. Which option is aligned with this outcome?

  1. Negative prompting
  2. Prompt A/B testing
  3. Prompt delimiters
  4. Amazon Bedrock Prompt Management

Explanation

Correct Answer

D:

This is the choice aligned with the industrial maintenance requirement to centrally manage and version reusable FM prompts: Bedrock Prompt Management supports prompt variants and managed lifecycle for supported workflows. A realistic pilot must still check the qualification that it is not a substitute for application-specific testing and permission management, especially where that could affect the acceptance criterion.

Incorrect Answers

A:

For the industrial maintenance objective to centrally manage and version reusable FM prompts, Negative prompting addresses a different task—tell an image or text generator what elements to avoid. Its operating mechanism differs from that of Amazon Bedrock Prompt Management: negative guidance identifies unwanted styles, topics or content traits; this does not satisfy the requested outcome.

B:

For the industrial maintenance objective to centrally manage and version reusable FM prompts, Prompt A/B testing addresses a different task—compare two instruction designs on the same representative workload. Its operating mechanism differs from that of Amazon Bedrock Prompt Management: a controlled evaluation measures changes in accuracy, safety and format adherence; this does not satisfy the requested outcome.

C:

For the industrial maintenance objective to centrally manage and version reusable FM prompts, Prompt delimiters addresses a different task—separate untrusted customer text from the application's own instructions. Its operating mechanism differs from that of Amazon Bedrock Prompt Management: explicit boundaries distinguish source content from the task instructions; this does not satisfy the requested outcome.

Question 627

A stakeholder in banking operations asks how to ask a model to complete a standard task without sample answers. Which of the listed approaches fits that request?

  1. Few-shot prompting
  2. Prompt delimiters
  3. Zero-shot prompting
  4. Task-specific instructions

Explanation

Correct Answer

C:

This is the choice aligned with the banking operations requirement to ask a model to complete a standard task without sample answers: zero-shot prompting relies on the model's existing general capabilities. A realistic pilot must still check the qualification that specialized formats may benefit from examples or stricter output specifications, especially where that could affect the acceptance criterion.

Incorrect Answers

A:

For the banking operations objective to ask a model to complete a standard task without sample answers, Few-shot prompting addresses a different task—teach a desired labeling convention using several representative examples. Its operating mechanism differs from that of Zero-shot prompting: several input-output examples make the intended pattern explicit without training; this does not satisfy the requested outcome.

B:

For the banking operations objective to ask a model to complete a standard task without sample answers, Prompt delimiters addresses a different task—separate untrusted customer text from the application's own instructions. Its operating mechanism differs from that of Zero-shot prompting: explicit boundaries distinguish source content from the task instructions; this does not satisfy the requested outcome.

D:

For the banking operations objective to ask a model to complete a standard task without sample answers, Task-specific instructions addresses a different task—improve an ambiguous request by stating the task and success criteria. Its operating mechanism differs from that of Zero-shot prompting: specific instructions describe what to do, for whom and in what output form; this does not satisfy the requested outcome.

Question 628

A subscription services proof of concept is intended to direct the assistant to use only retrieved passages for a policy answer. Which concept is most applicable to the stated goal?

  1. Grounded answering instructions
  2. Zero-shot prompting
  3. Prompt poisoning
  4. Prompt evaluation dataset

Explanation

Correct Answer

A:

This is the choice aligned with the subscription services requirement to direct the assistant to use only retrieved passages for a policy answer: grounding instructions establish evidence boundaries and a fallback when sources are missing. A realistic pilot must still check the qualification that without retrieval and checking, a command to cite sources may generate invented citations, especially where that could affect the acceptance criterion.

Incorrect Answers

B:

For the subscription services objective to direct the assistant to use only retrieved passages for a policy answer, Zero-shot prompting addresses a different task—ask a model to complete a standard task without sample answers. Its operating mechanism differs from that of Grounded answering instructions: zero-shot prompting relies on the model's existing general capabilities; this does not satisfy the requested outcome.

C:

For the subscription services objective to direct the assistant to use only retrieved passages for a policy answer, Prompt poisoning addresses a different task—prevent compromised retrieval material from manipulating the answer. Its operating mechanism differs from that of Grounded answering instructions: poisoned source content is deliberately planted to influence downstream model output; this does not satisfy the requested outcome.

D:

For the subscription services objective to direct the assistant to use only retrieved passages for a policy answer, Prompt evaluation dataset addresses a different task—build a repeatable collection of normal and adversarial test prompts. Its operating mechanism differs from that of Grounded answering instructions: a representative evaluation set checks regressions in task success and safety; this does not satisfy the requested outcome.

Question 629

To teach a desired labeling convention using several representative examples, a telecommunications project considers Few-shot prompting. Which statement correctly describes its role?

  1. Several input-output examples make the intended pattern explicit without training.
  2. A controlled evaluation measures changes in accuracy, safety and format adherence.
  3. A template separates stable instructions from variables inserted for each request.
  4. Versioned prompts preserve tested text and configuration across deployments.

Explanation

Correct Answer

A:

For the proposed telecommunications use, the key operating principle is that several input-output examples make the intended pattern explicit without training; that mechanism is what enables Few-shot prompting to teach a desired labeling convention using several representative examples. The technical limitation—too many examples add token cost and can bias the model toward narrow patterns—needs a separate check.

Incorrect Answers

B:

In this telecommunications question about Few-shot prompting, the statement instead characterizes Prompt A/B testing: a controlled evaluation measures changes in accuracy, safety and format adherence. That can support a goal to compare two instruction designs on the same representative workload, but does not explain the operation required for Few-shot prompting to teach a desired labeling convention using several representative examples.

C:

In this telecommunications question about Few-shot prompting, the statement instead characterizes Structured prompt template: a template separates stable instructions from variables inserted for each request. That can support a goal to reuse a tested set of instructions with different customer details, but does not explain the operation required for Few-shot prompting to teach a desired labeling convention using several representative examples.

D:

In this telecommunications question about Few-shot prompting, the statement instead characterizes Prompt versioning: versioned prompts preserve tested text and configuration across deployments. That can support a goal to reproduce and roll back a prompt that behaved better last week, but does not explain the operation required for Few-shot prompting to teach a desired labeling convention using several representative examples.

Question 630

The online retail team is reviewing how Prompt injection recognition works in a plan to detect instructions hidden in retrieved web pages that attempt to override the user task. Which statement gives the correct mechanism?

  1. Versioned prompts preserve tested text and configuration across deployments.
  2. Breaking a complex task into smaller verifiable steps can improve task execution.
  3. Specific instructions describe what to do, for whom and in what output form.
  4. Prompt injection abuses lower-trust content to redirect a model's behavior.

Explanation

Correct Answer

D:

For the proposed online retail use, the key operating principle is that prompt injection abuses lower-trust content to redirect a model's behavior; that mechanism is what enables Prompt injection recognition to detect instructions hidden in retrieved web pages that attempt to override the user task. The technical limitation—treating a web snippet as trusted system policy creates an unsafe authority boundary—needs a separate check.

Incorrect Answers

A:

In this online retail question about Prompt injection recognition, the statement instead characterizes Prompt versioning: versioned prompts preserve tested text and configuration across deployments. That can support a goal to reproduce and roll back a prompt that behaved better last week, but does not explain the operation required for Prompt injection recognition to detect instructions hidden in retrieved web pages that attempt to override the user task.

B:

In this online retail question about Prompt injection recognition, the statement instead characterizes Chain-of-thought-style decomposition: breaking a complex task into smaller verifiable steps can improve task execution. That can support a goal to improve a multi-step reasoning task through intermediate checks or subquestions, but does not explain the operation required for Prompt injection recognition to detect instructions hidden in retrieved web pages that attempt to override the user task.

C:

In this online retail question about Prompt injection recognition, the statement instead characterizes Task-specific instructions: specific instructions describe what to do, for whom and in what output form. That can support a goal to improve an ambiguous request by stating the task and success criteria, but does not explain the operation required for Prompt injection recognition to detect instructions hidden in retrieved web pages that attempt to override the user task.

AWS Certified AI Practitioner AIF-C01 Practice Tests