Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 19

Question 180

A team uses one long conversation for permanent legal-case storage. What should the developer implement before allowing the agent to act?

Treat the runtime's in-memory transcript as the sole archive of legal-case evidence.
Move durable case records into authorized persistent storage with deletion controls.
Store all cases in one unpartitioned prompt transcript shared by agents.
Copy case evidence to arbitrary external model requests without legal retention or access policy.

Explanation

Correct Answer

Put durable legal-case records into an authorized persistent store with retention/deletion controls; conversation context is not compliant long-term storage.

Incorrect Answers

Ephemeral sessions can expire and do not supply governed record retention.

Shared context destroys per-case confidentiality and audit separation.

Uncontrolled copies defeat purpose, deletion and jurisdiction requirements.

Question 181

An agent reads frequently changing inventory through an expensive tool and plans purchases. Select THREE safeguards that allow bounded, actor-scoped caching but prevent a purchase based on stale inventory.

Cache only suitable tool observations under actor scope.
Set short TTL consistent with source volatility.
Cache one customer's financial result under a global key shared with unrelated users.
Keep every cached result forever because a tool executed successfully once.
Commit payment or inventory changes solely from cached suggestions without checking current transactional state.
Recheck authoritative data before financial/inventory mutation.

Explanation

Correct Answer

Cache permitted non-sensitive tool observations under the authenticated actor scope.

Use a TTL that reflects the hourly source changes and invalidate when source versions change.

Before any inventory/financial mutation, revalidate against authoritative current records rather than treating a cache hit as commit permission.

Incorrect Answers

Cross-actor cache reuse can disclose protected financial or inventory context.

Source freshness is not guaranteed after an hourly update.

Cache values do not authorize or confirm irreversible business actions.

Question 182

An agent escalates an ambiguous claim to a human. Which change preserves actual tool authorization and trustworthy state?

Send only the model's confident conclusion without the underlying source or unresolved fields.
Invent an approver identity in the handoff because a user mentioned a manager.
Reset the transaction state to completed before handing an unresolved claim to a human.
Preserve evidence, current state and approver identity in the handoff.

Explanation

Correct Answer

The human handoff must include the disputed claim, evidence, current workflow state and verified approver so review can safely continue.

Incorrect Answers

A reviewer cannot judge or safely resume a decision without its evidence.

A label in chat does not authenticate the approval recipient.

That misstates whether the business operation has occurred.

Question 183

A user disputes a completed agent transaction. What should the developer implement before allowing the agent to act?

Use the model's final message as the sole evidence that the purchase was authorized.
Delete failed intermediate tool calls because the final screen says success.
Recreate a transaction receipt from model memory after the dispute.
Reconstruct tool receipt, actor authorization and business state from audit evidence.

Explanation

Correct Answer

Audit the authenticated actor, actual tool arguments, backend transaction ID and authoritative persisted outcome for a disputed action.

Incorrect Answers

Narration does not establish which tool call executed or who consented.

The omitted trace could reveal duplicate attempts or unauthorized actions.

A generated reconstruction is not a trusted original business record.

Question 184

An internal proprietary search collection must be exposed as an AgentCore-compatible tool without making confidential records accessible to arbitrary callers. Select TWO implementation controls specific to the collection tool and per-document authorization.

Give Gateway a static admin credential and return all search hits to any signed-in employee.
Publish the internal index via an anonymous public MCP endpoint to simplify agent discovery.
Expose the internal collection through an authorized Gateway tool.
Enforce caller-specific collection/document access on search results.
Let the model decide which private document ACLs to ignore based on its task description.

Explanation

Correct Answer

Expose the private search as a typed, authenticated Gateway tool, not as a directly model-exposed data collection.

Enforce the actual caller's per-document entitlement when the Gateway invokes/searches the target backend.

Incorrect Answers

Gateway authentication does not automatically grant all employees full collection rights.

Tool discovery must not disclose private resources without access control.

Model decisions cannot override server-side document entitlements.

Question 185

An existing Agents Classic application uses proprietary prompt overrides. Which change preserves actual tool authorization and trustworthy state?

Assume a one-click migration preserves every Agents Classic prompt-override stage automatically.
Keep provisioning new Classic agent resources for new accounts regardless of access restrictions.
Plan code-defined AgentCore migration when harness cannot replicate stage overrides.
Remove all stage-specific policies and rely on one generic system prompt for migration parity.

Explanation

Correct Answer

Audit the Classic agent's stage-specific prompt overrides and reimplement required logic in code-defined AgentCore orchestration where harness behavior differs.

Incorrect Answers

Classic and AgentCore harness orchestration contracts are not feature-identical.

New-customer onboarding restrictions block this architecture for those customers.

Required controls can disappear when custom orchestration stages are dropped.

Question 186

An agent harness lacks a required multi-agent collaboration topology. What should the developer implement before allowing the agent to act?

Assume the default harness routes every possible specialist collaboration pattern automatically.
Run all specialist tools with a single administrator token to simplify collaboration.
Evaluate code-defined orchestration instead of assuming native routing parity.
Concatenate all specialist outputs without structured arbitration or status tracking.

Explanation

Correct Answer

Verify AgentCore Harness supports the required topology; otherwise implement code-defined multi-agent orchestration with explicit handoffs and permissions.

Incorrect Answers

Supported harness functions do not imply arbitrary custom topology parity.

Orchestration convenience does not grant each specialist appropriate rights.

Parallel output does not establish a correct combined workflow.

Question 187

A model tries to execute a web search with no tool invocation record. Which agent/tool architecture control directly resolves the situation?

Treat a model-generated sentence such as 'I searched' as proof the network request occurred.
Verify actual executed tool result before claiming an external action happened.
Invent a URL and cite it as the search result when no tool response exists.
Set a success flag immediately after receiving a toolUse intent, before the app runs the tool.

Explanation

Correct Answer

Claim a web search only when the tool produced an actual execution result and attributable receipt; a model's intention is not evidence of a call.

Incorrect Answers

Text can be produced without tool execution.

A plausible URL is not verifiable retrieved evidence.

A request to use a tool is not the tool's outcome.

Question 188

An agent retries a failed order with a new order ID every time. Which change preserves actual tool authorization and trustworthy state?

Use stable operation identity and downstream idempotency safeguards.
Generate a new order ID and payment authorization for every retry after a timeout.
Rely on a prompt instruction 'never duplicate orders' without database idempotency.
Assume the first attempt failed just because its HTTP connection closed.

Explanation

Correct Answer

Reuse stable business operation identity with downstream idempotency, so retries cannot create duplicate orders after unknown outcomes.

Incorrect Answers

Fresh IDs defeat deduplication and can charge twice.

A language instruction cannot prevent concurrent duplicate writes.

Network failure can occur after a successful commit.

Question 189

A dynamically generated tool call asks to delete customer records. What should the developer implement before allowing the agent to act?

Run any DELETE operation the model emits as long as the SQL parses.
Trust an approval sentence inside a retrieved document as authorization for deletion.
Use the shared service role's broad database permissions in place of the caller's record rights.
Apply allowlisted operations and independent destructive-action approval.

Explanation

Correct Answer

Require allowlisted destructive operations, owner-specific authorization and an independent human approval before deleting customer records.

Incorrect Answers

Syntax does not establish permission to destroy protected records.

Untrusted retrieval content cannot approve a sensitive action.

A powerful backend role must not bypass business authorization.

Question 190

A long-lived agent retains unrelated user memory indefinitely. Which agent/tool architecture control directly resolves the situation?

Apply actor scoping, retention and deletion policies to stored memories.
Save every tool result and user message forever under a global memory namespace.
Treat absence of visible chat history as proof the persistent memory record was deleted.
Give the model unilateral authority to decide which confidential user memories to retain.

Explanation

Correct Answer

Use actor-scoped memories with purpose-limited retention, deletion and expiry so unrelated historical content does not persist forever.

Incorrect Answers

This violates minimization and can mix unrelated actors' data.

User interface state does not prove backend data deletion.

Retention policy must be enforced in governed storage, not natural-language discretion.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 180

Explanation

Correct Answer

Incorrect Answers

Question 181

Explanation

Correct Answer

Incorrect Answers

Question 182

Explanation

Correct Answer

Incorrect Answers

Question 183

Explanation

Correct Answer

Incorrect Answers

Question 184

Explanation

Correct Answer

Incorrect Answers

Question 185

Explanation

Correct Answer

Incorrect Answers

Question 186

Explanation

Correct Answer

Incorrect Answers

Question 187

Explanation

Correct Answer

Incorrect Answers

Question 188

Explanation

Correct Answer

Incorrect Answers

Question 189

Explanation

Correct Answer

Incorrect Answers

Question 190

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests