Topic 10 Practice Test 2 covers Data Security, Privacy, Network Isolation, PII Protection, Masking, and Retention in Domain 3 (Task 3.2). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
The application itself must have no public IP addresses and must reach supported AWS services without NAT internet egress. What should the team implement? Choose ONE.
- VPC endpoint for private service access
- Private subnet without public IP plus VPC endpoints
- TLS for data in transit
- Security groups for endpoint access
Correct Answer: B
Correct Answer
Answer B is correct because Place the application in private subnets without public IP addresses and use the appropriate VPC endpoints for supported AWS services. Endpoint traffic stays on private AWS connectivity, so the workload does not need an internet gateway or NAT path merely to reach those services. That directly satisfies both isolation requirements.
Incorrect Answers
Answer A is incorrect because VPC endpoint for private service access — in the current architecture, keeps supported AWS service traffic, in the described flow, on private connectivity without requiring, in the described flow, public internet traversal. Applies where a protected GenAI environment must, in this case, avoid public network paths. Target outcome: isolate the application tier from public addressing and internet egress. VPC endpoint for private service access misses that outcome.
Answer C is incorrect because TLS for data in transit — in the current architecture, protects data while it moves, in the described flow, between clients and services using, in the described flow, encrypted transport. Applies where the threat concerns interception of, in this case, traffic in transit. Target outcome: isolate the application tier from public addressing and internet egress. TLS for data in transit misses that outcome.
Answer D is incorrect because Security groups for endpoint access — under these constraints, restrict network flows to the, at this layer, intended sources and destinations around, at this layer, VPC resources or endpoints. Choose it when private connectivity exists but network-level, under these constraints, reachability still needs narrowing. Requirement: isolate the application tier from public addressing and internet egress. Security groups for endpoint access is not the direct fit.
Question 2
A model-generated response may accidentally echo an account number from retrieved context; responses must be checked before delivery. Which option most directly addresses the requirement? Choose ONE.
- Bedrock Guardrails sensitive-information filter
- Amazon Comprehend PII detection
- Post-output PII filter
- Pre-prompt PII redaction
Correct Answer: C
Correct Answer
Answer C is correct because Post-output PII filter — at this decision point, checks generated content for sensitive, within this design, information before returning it to, within this design, a consumer. Typical use: the risk is accidental disclosure, in the described flow, in model output. Goal: detect or remove sensitive information at the output boundary. Post-output PII filter provides that behavior.
Incorrect Answers
Answer A is incorrect because Bedrock Guardrails sensitive-information filter — in the current architecture, can detect and block or, in the described flow, mask configured sensitive information in, in the described flow, model interactions. Applies where sensitive data must be filtered, in this case, at the GenAI interaction boundary. Target outcome: detect or remove sensitive information at the output boundary. Bedrock Guardrails sensitive-information filter misses that outcome.
Answer B is incorrect because Amazon Comprehend PII detection — for the described workload, detects PII entities in text, for the stated need, so an application can identify, for the stated need, sensitive text before or during, for the stated need, processing. Best suited when the application must find PII, for this decision, in text content programmatically. Deciding factor: detect or remove sensitive information at the output boundary. Amazon Comprehend PII detection addresses something else.
Answer D is incorrect because Pre-prompt PII redaction — for this design, removes detected PII from the, for this workload, prompt before model invocation so, for this workload, unnecessary sensitive values never reach, for this workload, the model. Fits when the model can complete the, for the stated need, task without raw PII. Constraint: detect or remove sensitive information at the output boundary. Pre-prompt PII redaction targets another layer.
Question 3
Prompts and outputs are allowed to exist for seven days for incident investigation and then must be deleted. Which design best satisfies this requirement? Choose ONE.
- Data minimization
- Anonymization
- Short retention plus explicit deletion
- S3 Lifecycle retention policy
Correct Answer: C
Correct Answer
Answer C is correct because Configure a seven-day retention period and an explicit deletion control for prompts and outputs. That keeps the artifacts available for the approved incident-investigation window and removes them when the obligation ends. The design directly enforces the required maximum lifetime instead of retaining data indefinitely after its operational purpose expires.
Incorrect Answers
Answer A is incorrect because Data minimization — under these constraints, limits collection and prompt context, at this layer, to only the data required, at this layer, for the stated purpose. Choose it when privacy risk should be reduced, under these constraints, before applying downstream controls. Requirement: enforce the approved short lifetime of interaction artifacts. Data minimization is not the direct fit.
Answer B is incorrect because Anonymization — for the described workload, removes or transforms identifying attributes, for the stated need, so the resulting data is, for the stated need, not intended to be linked, for the stated need, back to an individual. Best suited when the use case does not, for this decision, require re-identification or stable subject, for this decision, linkage. Deciding factor: enforce the approved short lifetime of interaction artifacts. Anonymization addresses something else.
Answer D is incorrect because S3 Lifecycle retention policy — at this decision point, transitions or expires S3 objects, within this design, according to age-based lifecycle rules. Typical use: stored prompts, outputs, or evidence, in the described flow, must be deleted or tiered, in the described flow, after a defined retention period. Goal: enforce the approved short lifetime of interaction artifacts. S3 Lifecycle retention policy does not provide it.
Question 4
A model-invocation role should not also carry permission to read the HR data lake; the retrieval component should have a separate permission set. What is the most appropriate implementation choice? Choose ONE.
- Lake Formation fine-grained access
- Cross-account IAM role
- IAM resource-scoped policy
- Separate data-access role from model role
Correct Answer: D
Correct Answer
Answer D is correct because Separate data-access role from model role — in this scenario, uses distinct roles so model, in this case, invocation permissions and sensitive-data permissions, in this case, can be governed independently. Useful when separation of duties and blast-radius, at this layer, reduction are required. Need: separate model permissions from sensitive-data permissions. Separate data-access role from model role matches it directly.
Incorrect Answers
Answer A is incorrect because Lake Formation fine-grained access — at this decision point, controls governed access to data-lake, within this design, resources at granular database, table,, within this design, column, or related policy boundaries. Typical use: different GenAI consumers need controlled, in the described flow, access to shared lake data. Goal: separate model permissions from sensitive-data permissions. Lake Formation fine-grained access does not provide it.
Answer B is incorrect because Cross-account IAM role — for the described workload, provides temporary, scoped cross-account access, for the stated need, without sharing long-lived user credentials. Best suited when a workload in one AWS, for this decision, account must access approved resources, for this decision, in another. Deciding factor: separate model permissions from sensitive-data permissions. Cross-account IAM role addresses something else.
Answer C is incorrect because IAM resource-scoped policy — for this operational need, grants only the required actions, under these constraints, against the specific model, data,, under these constraints, or application resources that need, under these constraints, access. Architects use it when the requirement is authorization minimization, for this workload, rather than network isolation. Key need: separate model permissions from sensitive-data permissions. IAM resource-scoped policy does not satisfy it.
Question 5
A source bucket should allow the RAG ingestion role to read only /approved/ and deny unrelated prefixes. Which approach should the developer use? Choose ONE.
- S3 Block Public Access
- IAM resource-scoped policy
- Lake Formation fine-grained access
- Least-privilege bucket policy
Correct Answer: D
Correct Answer
Answer D is correct because Least-privilege bucket policy — at this decision point, restricts S3 object access to, within this design, the exact principals, prefixes, conditions,, within this design, or transport requirements needed. Typical use: the risk is overly broad, in the described flow, access to source documents. Goal: limit S3 source access to the required prefix. Least-privilege bucket policy provides that behavior.
Incorrect Answers
Answer A is incorrect because S3 Block Public Access — under these constraints, prevents public exposure paths for, at this layer, S3 buckets and objects at, at this layer, the account or bucket level. Choose it when the key risk is accidental, under these constraints, public access to S3 data. Requirement: limit S3 source access to the required prefix. S3 Block Public Access is not the direct fit.
Answer B is incorrect because IAM resource-scoped policy — within this implementation, grants only the required actions, for this decision, against the specific model, data,, for this decision, or application resources that need, for this decision, access. Strong fit when the requirement is authorization minimization, within this design, rather than network isolation. Operational need: limit S3 source access to the required prefix. IAM resource-scoped policy is secondary here.
Answer C is incorrect because Lake Formation fine-grained access — under these constraints, controls governed access to data-lake, at this layer, resources at granular database, table,, at this layer, column, or related policy boundaries. Choose it when different GenAI consumers need controlled, under these constraints, access to shared lake data. Requirement: limit S3 source access to the required prefix. Lake Formation fine-grained access is not the direct fit.
Question 6
Documents may contain PII that should never enter the vector index. The system should detect and label sensitive content before indexing. What should the team implement? Choose ONE.
- PII classification before indexing
- Post-output PII filter
- Amazon Comprehend PII detection
- Amazon Macie sensitive-data discovery
Correct Answer: A
Correct Answer
Answer A is correct because PII classification before indexing — for this design, detects and labels sensitive content, for this workload, before it enters a vector, for this workload, or retrieval index. Fits when retrieval should not expose sensitive, for the stated need, chunks that should never have, for the stated need, been indexed. Constraint: stop sensitive chunks before they become retrievable vectors. PII classification before indexing is the closest fit.
Incorrect Answers
Answer B is incorrect because Post-output PII filter — under these constraints, checks generated content for sensitive, at this layer, information before returning it to, at this layer, a consumer. Choose it when the risk is accidental disclosure, under these constraints, in model output. Requirement: stop sensitive chunks before they become retrievable vectors. Post-output PII filter is not the direct fit.
Answer C is incorrect because Amazon Comprehend PII detection — in the current architecture, detects PII entities in text, in the described flow, so an application can identify, in the described flow, sensitive text before or during, in the described flow, processing. Applies where the application must find PII, in this case, in text content programmatically. Target outcome: stop sensitive chunks before they become retrievable vectors. Amazon Comprehend PII detection misses that outcome.
Answer D is incorrect because Amazon Macie sensitive-data discovery — for the described workload, discovers and classifies sensitive data, for the stated need, such as PII in Amazon, for the stated need, S3 using managed data-security capabilities. Best suited when the organization must locate sensitive, for this decision, information stored in S3 at, for this decision, scale. Deciding factor: stop sensitive chunks before they become retrievable vectors. Amazon Macie sensitive-data discovery addresses something else.
Question 7
A data lake hosts HR, finance, and product data. A GenAI analyst should see only approved columns from product tables. Which option most directly addresses the requirement? Choose ONE.
- Data masking before model invocation
- Lake Formation fine-grained access
- Least-privilege bucket policy
- IAM resource-scoped policy
Correct Answer: B
Correct Answer
Answer B is correct because Lake Formation fine-grained access — for the described workload, controls governed access to data-lake, for the stated need, resources at granular database, table,, for the stated need, column, or related policy boundaries. Best suited when different GenAI consumers need controlled, for this decision, access to shared lake data. Deciding factor: govern table and column access centrally. Lake Formation fine-grained access fits this case.
Incorrect Answers
Answer A is incorrect because Data masking before model invocation — for this operational need, replaces sensitive values with masked, under these constraints, representations before they are sent, under these constraints, to the model. Architects use it when the model does not need, for this workload, the original sensitive value to, for this workload, perform the task. Key need: govern table and column access centrally. Data masking before model invocation does not satisfy it.
Answer C is incorrect because Least-privilege bucket policy — under these constraints, restricts S3 object access to, at this layer, the exact principals, prefixes, conditions,, at this layer, or transport requirements needed. Choose it when the risk is overly broad, under these constraints, access to source documents. Requirement: govern table and column access centrally. Least-privilege bucket policy is not the direct fit.
Answer D is incorrect because IAM resource-scoped policy — in this scenario, grants only the required actions, in this case, against the specific model, data,, in this case, or application resources that need, in this case, access. Appropriate if the requirement is authorization minimization, at this layer, rather than network isolation. The stem requires govern table and column access centrally. IAM resource-scoped policy serves another purpose.
Question 8
A batch of S3 documents needs managed discovery to determine which objects contain sensitive data before a GenAI migration. Which design best satisfies this requirement? Choose ONE.
- CloudWatch access monitoring
- Amazon Comprehend PII detection
- Amazon Macie sensitive-data discovery
- PII classification before indexing
Correct Answer: C
Correct Answer
Answer C is correct because Amazon Macie sensitive-data discovery — in the current architecture, discovers and classifies sensitive data, in the described flow, such as PII in Amazon, in the described flow, S3 using managed data-security capabilities. Applies where the organization must locate sensitive, in this case, information stored in S3 at, in this case, scale. Target outcome: scan S3 content for sensitive-data discovery. Amazon Macie sensitive-data discovery supplies it directly.
Incorrect Answers
Answer A is incorrect because CloudWatch access monitoring — at this decision point, collects operational telemetry and alarms, within this design, so suspicious or unexpected access, within this design, behavior can be detected. Typical use: security teams need monitoring rather, in the described flow, than a preventive data transformation. Goal: scan S3 content for sensitive-data discovery. CloudWatch access monitoring does not provide it.
Answer B is incorrect because Amazon Comprehend PII detection — for this operational need, detects PII entities in text, under these constraints, so an application can identify, under these constraints, sensitive text before or during, under these constraints, processing. Architects use it when the application must find PII, for this workload, in text content programmatically. Key need: scan S3 content for sensitive-data discovery. Amazon Comprehend PII detection does not satisfy it.
Answer D is incorrect because PII classification before indexing — at this decision point, detects and labels sensitive content, within this design, before it enters a vector, within this design, or retrieval index. Typical use: retrieval should not expose sensitive, in the described flow, chunks that should never have, in the described flow, been indexed. Goal: scan S3 content for sensitive-data discovery. PII classification before indexing does not provide it.
Question 9
An application processes a single user’s text and needs entity-level PII detection immediately before constructing the prompt. What is the most appropriate implementation choice? Choose ONE.
- Data minimization
- Amazon Macie sensitive-data discovery
- Bedrock Guardrails sensitive-information filter
- Amazon Comprehend PII detection
Correct Answer: D
Correct Answer
Answer D is correct because Amazon Comprehend PII detection — within this implementation, detects PII entities in text, for this decision, so an application can identify, for this decision, sensitive text before or during, for this decision, processing. Strong fit when the application must find PII, within this design, in text content programmatically. Operational need: identify PII entities inside application text. Amazon Comprehend PII detection is appropriate here.
Incorrect Answers
Answer A is incorrect because Data minimization — for the described workload, limits collection and prompt context, for the stated need, to only the data required, for the stated need, for the stated purpose. Best suited when privacy risk should be reduced, for this decision, before applying downstream controls. Deciding factor: identify PII entities inside application text. Data minimization addresses something else.
Answer B is incorrect because Amazon Macie sensitive-data discovery — for this operational need, discovers and classifies sensitive data, under these constraints, such as PII in Amazon, under these constraints, S3 using managed data-security capabilities. Architects use it when the organization must locate sensitive, for this workload, information stored in S3 at, for this workload, scale. Key need: identify PII entities inside application text. Amazon Macie sensitive-data discovery does not satisfy it.
Answer C is incorrect because Bedrock Guardrails sensitive-information filter — for this operational need, can detect and block or, under these constraints, mask configured sensitive information in, under these constraints, model interactions. Architects use it when sensitive data must be filtered, for this workload, at the GenAI interaction boundary. Key need: identify PII entities inside application text. Bedrock Guardrails sensitive-information filter does not satisfy it.
Question 10
An organization wants the model interface itself to block configured categories of sensitive information in prompts and responses. Which approach should the developer use? Choose ONE.
- Bedrock Guardrails sensitive-information filter
- Post-output PII filter
- AWS KMS encryption
- Pre-prompt PII redaction
Correct Answer: A
Correct Answer
Answer A is correct because Bedrock Guardrails sensitive-information filter — within this implementation, can detect and block or, for this decision, mask configured sensitive information in, for this decision, model interactions. Strong fit when sensitive data must be filtered, within this design, at the GenAI interaction boundary. Operational need: enforce sensitive-information handling at the Bedrock interaction layer. Bedrock Guardrails sensitive-information filter is appropriate here.
Incorrect Answers
Answer B is incorrect because Post-output PII filter — for the described workload, checks generated content for sensitive, for the stated need, information before returning it to, for the stated need, a consumer. Best suited when the risk is accidental disclosure, for this decision, in model output. Deciding factor: enforce sensitive-information handling at the Bedrock interaction layer. Post-output PII filter addresses something else.
Answer C is incorrect because AWS KMS encryption — in this scenario, uses customer-managed or AWS managed, in this case, cryptographic keys to protect supported, in this case, data at rest and control, in this case, key usage. Appropriate if the primary control is encryption, at this layer, and key governance for stored, at this layer, data. The stem requires enforce sensitive-information handling at the Bedrock interaction layer. AWS KMS encryption serves another purpose.
Answer D is incorrect because Pre-prompt PII redaction — at this decision point, removes detected PII from the, within this design, prompt before model invocation so, within this design, unnecessary sensitive values never reach, within this design, the model. Typical use: the model can complete the, in the described flow, task without raw PII. Goal: enforce sensitive-information handling at the Bedrock interaction layer. Pre-prompt PII redaction does not provide it.
Question 11
A pseudonymous customer token must remain stable across sessions so analysts can measure repeat usage without seeing the original email. What should the team implement? Choose ONE.
- Pseudonymization/tokenization
- Data minimization
- Anonymization
- Data masking before model invocation
Correct Answer: A
Correct Answer
Answer A is correct because Pseudonymization/tokenization — under these constraints, replaces direct identifiers with controlled, at this layer, tokens so records remain linkable, at this layer, without exposing the original identifier. Choose it when analytics needs stable linkage while, under these constraints, direct identifiers remain protected. Requirement: retain linkability while protecting direct identity. Pseudonymization/tokenization addresses it precisely.
Incorrect Answers
Answer B is incorrect because Data minimization — in the current architecture, limits collection and prompt context, in the described flow, to only the data required, in the described flow, for the stated purpose. Applies where privacy risk should be reduced, in this case, before applying downstream controls. Target outcome: retain linkability while protecting direct identity. Data minimization misses that outcome.
Answer C is incorrect because Anonymization — in the current architecture, removes or transforms identifying attributes, in the described flow, so the resulting data is, in the described flow, not intended to be linked, in the described flow, back to an individual. Applies where the use case does not, in this case, require re-identification or stable subject, in this case, linkage. Target outcome: retain linkability while protecting direct identity. Anonymization misses that outcome.
Answer D is incorrect because Data masking before model invocation — within this implementation, replaces sensitive values with masked, for this decision, representations before they are sent, for this decision, to the model. Strong fit when the model does not need, within this design, the original sensitive value to, within this design, perform the task. Operational need: retain linkability while protecting direct identity. Data masking before model invocation is secondary here.
Question 12
A training/evaluation export is going to an external research partner and there is no need to reconnect records to real users. Which option most directly addresses the requirement? Choose ONE.
- AWS KMS encryption
- Anonymization
- Short retention plus explicit deletion
- Pseudonymization/tokenization
Correct Answer: B
Correct Answer
Answer B is correct because Anonymization — for this operational need, removes or transforms identifying attributes, under these constraints, so the resulting data is, under these constraints, not intended to be linked, under these constraints, back to an individual. Architects use it when the use case does not, for this workload, require re-identification or stable subject, for this workload, linkage. Key need: remove identifying linkage before external research sharing. Anonymization aligns with it.
Incorrect Answers
Answer A is incorrect because AWS KMS encryption — for this design, uses customer-managed or AWS managed, for this workload, cryptographic keys to protect supported, for this workload, data at rest and control, for this workload, key usage. Supports designs where the primary control is encryption, for the stated need, and key governance for stored, for the stated need, data. The workload needs remove identifying linkage before external research sharing. AWS KMS encryption maps elsewhere.
Answer C is incorrect because Short retention and explicit deletion control how long data is stored, not whether an external recipient can link a record to a real person. Because the partner never needs re-identification, the privacy requirement is unlinkability before sharing. Anonymization addresses that requirement directly; retention policy is a separate lifecycle control.
Answer D is incorrect because Pseudonymization/tokenization — for the described workload, replaces direct identifiers with controlled, for the stated need, tokens so records remain linkable, for the stated need, without exposing the original identifier. Best suited when analytics needs stable linkage while, for this decision, direct identifiers remain protected. Deciding factor: remove identifying linkage before external research sharing. Pseudonymization/tokenization addresses something else.
Question 13
An application stores retrieved documents encrypted, but audit finds every developer can use the KMS key to decrypt them. Which design best satisfies this requirement? Choose ONE.
- S3 Block Public Access
- AWS KMS encryption
- IAM resource-scoped policy
- KMS key policy and grants
Correct Answer: D
Correct Answer
Answer D is correct because KMS key policy and grants — for this design, controls which principals can use, for this workload, a KMS key for cryptographic, for this workload, operations. Fits when encrypted data exists but unauthorized, for the stated need, key use must be prevented. Constraint: narrow cryptographic key usage to authorized principals. KMS key policy and grants is the closest fit.
Incorrect Answers
Answer A is incorrect because S3 Block Public Access — for the described workload, prevents public exposure paths for, for the stated need, S3 buckets and objects at, for the stated need, the account or bucket level. Best suited when the key risk is accidental, for this decision, public access to S3 data. Deciding factor: narrow cryptographic key usage to authorized principals. S3 Block Public Access addresses something else.
Answer B is incorrect because AWS KMS encryption protects supported data with cryptographic keys, but encryption alone does not correct overly broad permission to use the key. The finding is specifically that every developer can decrypt the documents. Restricting the KMS key policy and grants to authorized principals fixes that authorization problem at the key boundary.
Answer C is incorrect because IAM resource-scoped policy — for this design, grants only the required actions, for this workload, against the specific model, data,, for this workload, or application resources that need, for this workload, access. Supports designs where the requirement is authorization minimization, for the stated need, rather than network isolation. The workload needs narrow cryptographic key usage to authorized principals. IAM resource-scoped policy maps elsewhere.
Question 14
A third-party API credential is currently embedded in a prompt template repository. What is the most appropriate implementation choice? Choose ONE.
- Data masking before model invocation
- AWS KMS encryption
- Secrets Manager for credentials
- IAM resource-scoped policy
Correct Answer: C
Correct Answer
Answer C is correct because Secrets Manager for credentials — under these constraints, stores and retrieves application secrets, at this layer, without hardcoding them into prompts,, at this layer, code, or configuration files. Choose it when the application must protect rotating, under these constraints, credentials used by integrations. Requirement: remove reusable secrets from code and prompt assets. Secrets Manager for credentials addresses it precisely.
Incorrect Answers
Answer A is incorrect because Data masking before model invocation — in this scenario, replaces sensitive values with masked, in this case, representations before they are sent, in this case, to the model. Appropriate if the model does not need, at this layer, the original sensitive value to, at this layer, perform the task. The stem requires remove reusable secrets from code and prompt assets. Data masking before model invocation serves another purpose.
Answer B is incorrect because AWS KMS encryption — under these constraints, uses customer-managed or AWS managed, at this layer, cryptographic keys to protect supported, at this layer, data at rest and control, at this layer, key usage. Makes sense when the primary control is encryption, under these constraints, and key governance for stored, under these constraints, data. The case depends on remove reusable secrets from code and prompt assets. AWS KMS encryption handles another concern.
Answer D is incorrect because IAM resource-scoped policy — at this decision point, grants only the required actions, within this design, against the specific model, data,, within this design, or application resources that need, within this design, access. Use it where the requirement is authorization minimization, in the described flow, rather than network isolation. The design requires remove reusable secrets from code and prompt assets. IAM resource-scoped policy is not the direct choice.
Question 15
A bucket is private but its policy grants read access to every role in the account. Only the ingestion role should read one prefix. Which approach should the developer use? Choose ONE.
- IAM resource-scoped policy
- Least-privilege bucket policy
- Lake Formation fine-grained access
- S3 Block Public Access
Correct Answer: B
Correct Answer
Answer B is correct because Least-privilege bucket policy — for the described workload, restricts S3 object access to, for the stated need, the exact principals, prefixes, conditions,, for the stated need, or transport requirements needed. Best suited when the risk is overly broad, for this decision, access to source documents. Deciding factor: reduce overly broad S3 authorization. Least-privilege bucket policy fits this case.
Incorrect Answers
Answer A is incorrect because IAM resource-scoped policy — under these constraints, grants only the required actions, at this layer, against the specific model, data,, at this layer, or application resources that need, at this layer, access. Makes sense when the requirement is authorization minimization, under these constraints, rather than network isolation. The case depends on reduce overly broad S3 authorization. IAM resource-scoped policy handles another concern.
Answer C is incorrect because Lake Formation fine-grained access — in the current architecture, controls governed access to data-lake, in the described flow, resources at granular database, table,, in the described flow, column, or related policy boundaries. Applies where different GenAI consumers need controlled, in this case, access to shared lake data. Target outcome: reduce overly broad S3 authorization. Lake Formation fine-grained access misses that outcome.
Answer D is incorrect because S3 Block Public Access — in the current architecture, prevents public exposure paths for, in the described flow, S3 buckets and objects at, in the described flow, the account or bucket level. Applies where the key risk is accidental, in this case, public access to S3 data. Target outcome: reduce overly broad S3 authorization. S3 Block Public Access misses that outcome.
Question 16
The application role can invoke Bedrock but should not list, read, or modify unrelated AWS resources. What should the team implement? Choose ONE.
- VPC endpoint for private service access
- IAM resource-scoped policy
- Tool allowlist
- Separate data-access role from model role
Correct Answer: B
Correct Answer
Answer B is correct because IAM resource-scoped policy — for the described workload, grants only the required actions, for the stated need, against the specific model, data,, for the stated need, or application resources that need, for the stated need, access. Intended when the requirement is authorization minimization, for this decision, rather than network isolation. The architecture needs scope permissions to the exact required API actions and resources. IAM resource-scoped policy fits best.
Incorrect Answers
Answer A is incorrect because VPC endpoint for private service access — for this operational need, keeps supported AWS service traffic, under these constraints, on private connectivity without requiring, under these constraints, public internet traversal. Architects use it when a protected GenAI environment must, for this workload, avoid public network paths. Key need: scope permissions to the exact required API actions and resources. VPC endpoint for private service access does not satisfy it.
Answer C is incorrect because Tool allowlist — for the described workload, restricts the agent to an, for the stated need, approved set of callable operations, for the stated need, instead of exposing every available, for the stated need, action. Intended when the risk boundary is which, for this decision, tools the model may invoke. The architecture needs scope permissions to the exact required API actions and resources. Tool allowlist is a poor fit.
Answer D is incorrect because Separate data-access role from model role — for this design, uses distinct roles so model, for this workload, invocation permissions and sensitive-data permissions, for this workload, can be governed independently. Fits when separation of duties and blast-radius, for the stated need, reduction are required. Constraint: scope permissions to the exact required API actions and resources. Separate data-access role from model role targets another layer.
Question 17
The security team wants to detect unusual increases in denied data-access attempts and alert operations. Which option most directly addresses the requirement? Choose ONE.
- CloudWatch access monitoring
- S3 Lifecycle retention policy
- Amazon Macie sensitive-data discovery
- AWS KMS encryption
Correct Answer: A
Correct Answer
Answer A is correct because CloudWatch access monitoring — under these constraints, collects operational telemetry and alarms, at this layer, so suspicious or unexpected access, at this layer, behavior can be detected. Choose it when security teams need monitoring rather, under these constraints, than a preventive data transformation. Requirement: observe and alert on access behavior. CloudWatch access monitoring addresses it precisely.
Incorrect Answers
Answer B is incorrect because S3 Lifecycle retention policy — under these constraints, transitions or expires S3 objects, at this layer, according to age-based lifecycle rules. Choose it when stored prompts, outputs, or evidence, under these constraints, must be deleted or tiered, under these constraints, after a defined retention period. Requirement: observe and alert on access behavior. S3 Lifecycle retention policy is not the direct fit.
Answer C is incorrect because Amazon Macie sensitive-data discovery — within this implementation, discovers and classifies sensitive data, for this decision, such as PII in Amazon, for this decision, S3 using managed data-security capabilities. Strong fit when the organization must locate sensitive, within this design, information stored in S3 at, within this design, scale. Operational need: observe and alert on access behavior. Amazon Macie sensitive-data discovery is secondary here.
Answer D is incorrect because AWS KMS encryption protects data at rest and governs use of cryptographic keys. It does not monitor denied data-access attempts or generate operational alerts when those denials spike. CloudWatch-based access monitoring is the direct fit because the requirement is behavioral visibility and alerting, not stronger ciphertext protection.
Question 18
The organization must ensure archived interaction logs cannot remain forever if a lifecycle job is missed. Which design best satisfies this requirement? Choose ONE.
- Short retention plus explicit deletion
- S3 Lifecycle retention policy
- Data minimization
- CloudWatch access monitoring
Correct Answer: A
Correct Answer
Answer A is correct because Use a short retention policy backed by explicit deletion so archived interaction logs are removed when the approved period ends. Making deletion an explicit privacy control avoids depending solely on a lifecycle job that could be missed or misconfigured. The requirement is bounded retention, not merely moving or encrypting old objects.
Incorrect Answers
Answer B is incorrect because S3 Lifecycle retention policy — for the described workload, transitions or expires S3 objects, for the stated need, according to age-based lifecycle rules. Best suited when stored prompts, outputs, or evidence, for this decision, must be deleted or tiered, for this decision, after a defined retention period. Deciding factor: make deletion an explicit privacy control at retention expiry. S3 Lifecycle retention policy addresses something else.
Answer C is incorrect because Data minimization — for this operational need, limits collection and prompt context, under these constraints, to only the data required, under these constraints, for the stated purpose. Architects use it when privacy risk should be reduced, for this workload, before applying downstream controls. Key need: make deletion an explicit privacy control at retention expiry. Data minimization does not satisfy it.
Answer D is incorrect because CloudWatch access monitoring — for the described workload, collects operational telemetry and alarms, for the stated need, so suspicious or unexpected access, for the stated need, behavior can be detected. Best suited when security teams need monitoring rather, for this decision, than a preventive data transformation. Deciding factor: make deletion an explicit privacy control at retention expiry. CloudWatch access monitoring addresses something else.
Question 19
A summarization task can work with last four digits of an account number rather than the full value. What is the most appropriate implementation choice? Choose ONE.
- TLS for data in transit
- Anonymization
- Data masking before model invocation
- Pseudonymization/tokenization
Correct Answer: C
Correct Answer
Answer C is correct because Mask the account number before model invocation, preserving only the last four digits that the summarization task actually needs. This follows data minimization while retaining the limited utility required by the prompt. Tokenization or full anonymization would be unnecessary when a simple partial representation safely satisfies the processing purpose.
Incorrect Answers
Answer A is incorrect because TLS for data in transit — for this operational need, protects data while it moves, under these constraints, between clients and services using, under these constraints, encrypted transport. Architects use it when the threat concerns interception of, for this workload, traffic in transit. Key need: preserve limited utility while hiding the unnecessary sensitive portion. TLS for data in transit does not satisfy it.
Answer B is incorrect because Anonymization — within this implementation, removes or transforms identifying attributes, for this decision, so the resulting data is, for this decision, not intended to be linked, for this decision, back to an individual. Strong fit when the use case does not, within this design, require re-identification or stable subject, within this design, linkage. Operational need: preserve limited utility while hiding the unnecessary sensitive portion. Anonymization is secondary here.
Answer D is incorrect because Pseudonymization/tokenization — in the current architecture, replaces direct identifiers with controlled, in the described flow, tokens so records remain linkable, in the described flow, without exposing the original identifier. Applies where analytics needs stable linkage while, in this case, direct identifiers remain protected. Target outcome: preserve limited utility while hiding the unnecessary sensitive portion. Pseudonymization/tokenization misses that outcome.
Question 20
A user-profile object contains 40 attributes, but only language and subscription tier are needed to answer the request. Which approach should the developer use? Choose ONE.
- Anonymization
- Data masking before model invocation
- Bedrock Guardrails sensitive-information filter
- Data minimization
Correct Answer: D
Correct Answer
Answer D is correct because Data minimization — within this implementation, limits collection and prompt context, for this decision, to only the data required, for this decision, for the stated purpose. Strong fit when privacy risk should be reduced, within this design, before applying downstream controls. Operational need: avoid transmitting unnecessary personal attributes. Data minimization is appropriate here.
Incorrect Answers
Answer A is incorrect because Anonymization — in this scenario, removes or transforms identifying attributes, in this case, so the resulting data is, in this case, not intended to be linked, in this case, back to an individual. Appropriate if the use case does not, at this layer, require re-identification or stable subject, at this layer, linkage. The stem requires avoid transmitting unnecessary personal attributes. Anonymization serves another purpose.
Answer B is incorrect because Data masking before model invocation — at this decision point, replaces sensitive values with masked, within this design, representations before they are sent, within this design, to the model. Use it where the model does not need, in the described flow, the original sensitive value to, in the described flow, perform the task. The design requires avoid transmitting unnecessary personal attributes. Data masking before model invocation is not the direct choice.
Answer C is incorrect because Bedrock Guardrails sensitive-information filter — in this scenario, can detect and block or, in this case, mask configured sensitive information in, in this case, model interactions. Appropriate if sensitive data must be filtered, at this layer, at the GenAI interaction boundary. The stem requires avoid transmitting unnecessary personal attributes. Bedrock Guardrails sensitive-information filter serves another purpose.
