McAfee Secure

Cisco 350-901 Bundle

Exam Code: 350-901

Exam Name Designing, Deploying, and Managing Network Automation Systems

Certification Provider: Cisco

Corresponding Certification: DevNet Professional

Cisco 350-901 Bundle $25.00

100% Cisco 350-901 Guaranteed Success With Testking Training Materials

Get Real 350-901 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

  • Questions & Answers

    350-901 Questions & Answers

    604 Questions & Answers

    The ultimate exam preparation tool, Questions & Answers mirrors the actual 350-901 exam while allowing you to hone and measure every aspect of your test-taking ability.

  • 350-901 Video Course

    350-901 Training Course

    106 Video Lectures

    350-901 Video Course is developed by Cisco Professionals to help you pass the 350-901 exam.

    Description

    This course will improve your knowledge and skills required to pass Designing, Deploying, and Managing Network Automation Systems exam.

Designing, Deploying, and Managing Network Automation Systems (350-901) Practice Tests

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Total Cost: $164.98
Bundle Price: $139.98

Purchase Individually

  • Questions & Answers

    Questions & Answers

    604 Questions

    $124.99
  • 350-901 Video Course

    Training Course

    106 Video Lectures

    $39.99

Cisco 350-901 Practice Test 50

Question 490

While investigating an AI recommendation accuracy study for fictional Sunrise Aeronautics, the team reports: A FastMCP tool hides a Cisco API timeout and fabricates an empty device interface list that the agent treats as a successful snapshot. What is the appropriate response to this network automation condition?

  1. Evaluate the agent on a representative labeled test set with difficult, adversarial and negative cases, measuring factual error and unsafe-action rates.
  2. Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data.
  3. Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values.
  4. Treat retrieved logs as untrusted data and enforce instruction hierarchy and independent tool permissions outside the model.

Explanation

Correct Answer

C:

Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values. This addresses the problem because tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. Confirm success using: Agent tool traces distinguish timeout, authorization and partial-data failures from genuine empty network results. Applied here, the relevant facts are: A FastMCP tool hides a Cisco API timeout and fabricates an empty device interface list that the agent treats as a successful snapshot. The question asks what is the appropriate response to this network automation condition.

Incorrect Answers

A:

This alternative is a response to another technical cause: high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. The reported case instead depends on: tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. The action that directly addresses it is “Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values”.

B:

The proposed action belongs to the mechanism: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. Here the stated fault follows a different mechanism: tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. The corrective choice is “Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values”.

D:

This alternative is a response to another technical cause: prompt injection attempts to convert lower-trust retrieved content into high-priority instructions across an AI-agent trust boundary. The reported case instead depends on: tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. The action that directly addresses it is “Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values”.

Question 491

At fictional Windward Maritime, during a GitLab deployment pipeline rehearsal, the operator observes: A GitLab network release uses an overly privileged runner that applies Terraform changes to unauthorized production workspaces. What should the team add to its network automation operating process?

  1. Add merge topology review and a dry-run diff approval for complex revert operations.
  2. Add dependency-impact review and target-branch CI validation to hotfix backport procedures.
  3. Use protected environments, least-privilege deploy tokens and progressive rollout approval conditions.
  4. Version lab topology artifacts and run automated readiness verification before configuration tests.

Explanation

Correct Answer

C:

Use protected environments, least-privilege deploy tokens and progressive rollout approval conditions. This prevents the recurrence because a deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. A matching immediate correction is: Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout. Applied here, the relevant facts are: A GitLab network release uses an overly privileged runner that applies Terraform changes to unauthorized production workspaces. The question asks what should the team add to its network automation operating process.

Incorrect Answers

A:

The proposed action belongs to the mechanism: reverting a merge requires an explicit mainline perspective so Git knows which parent is the retained baseline. Here the stated fault follows a different mechanism: a deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. The corrective choice is “Use protected environments, least-privilege deploy tokens and progressive rollout approval conditions”.

B:

The offered action targets: a cherry-picked commit does not automatically bring every ancestor or dependent change from its original branch. The case facts point instead to: a deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. Implement “Use protected environments, least-privilege deploy tokens and progressive rollout approval conditions” to address that specific condition.

D:

The proposed action belongs to the mechanism: a reproducible simulation needs controlled topology definitions and readiness gates to avoid timing-dependent false failures. Here the stated fault follows a different mechanism: a deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. The corrective choice is “Use protected environments, least-privilege deploy tokens and progressive rollout approval conditions”.

Question 492

At fictional Elmcrest Theatre, during a conversational network agent pilot, the operator observes: A generated network change says a link is down because one collector stream has missing data, while device operational state is healthy. Which targeted correction best addresses the network automation failure?

  1. Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules.
  2. Subject AI-suggested changes to schema and unit tests, simulated network execution, policy checks and human change approval.
  3. Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation.
  4. Bind current request context to fresh authorized inventory and explicit tenant or device selection before tool calls.

Explanation

Correct Answer

A:

Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules. This addresses the problem because aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Confirm success using: The validation crosschecks observed interface or VLAN state before accepting a proposed operation and flags missing telemetry as uncertainty. Applied here, the relevant facts are: A generated network change says a link is down because one collector stream has missing data, while device operational state is healthy. The question asks which targeted correction best addresses the network automation failure.

Incorrect Answers

B:

This alternative is a response to another technical cause: generative systems can propose plausible-looking but incorrect commands; independent executable validation is required for safe infrastructure changes. The reported case instead depends on: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. The action that directly addresses it is “Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules”.

C:

The proposed action belongs to the mechanism: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. Here the stated fault follows a different mechanism: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. The corrective choice is “Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules”.

D:

The offered action targets: conversation memory is not authoritative network state; stale context and scope leakage create wrong-target operational risk. The case facts point instead to: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Implement “Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules” to address that specific condition.

Question 493

At fictional Cobalt Ridge Bank, an acceptance review of a campus API automation rollout finds: A one-time diagnostic query is forced through the same slow infrastructure apply workflow used for persistent configuration changes. Which explanation best matches the actual broken dependency?

  1. Automation choice is driven by lifecycle and operational risk rather than by a single preferred programming tool.
  2. Terraform modules support repeated declarative resource patterns while typed inputs isolate deployment-specific configuration.
  3. YANG distinguishes configuration intent from read-only operational state; successful config writes do not establish runtime protocol health.
  4. Ansible's desired-state-aware network modules compare declared configuration with device state before determining required changes.

Explanation

Correct Answer

A:

Automation choice is driven by lifecycle and operational risk rather than by a single preferred programming tool. The corresponding repair is: Select the tool and deployment pattern by risk, state persistence, change scope, rollback capability and approval obligations. Acceptance should show: The pilot demonstrates appropriate staged changes for configuration updates and simple bounded queries for read-only diagnostics. Applied here, the relevant facts are: A one-time diagnostic query is forced through the same slow infrastructure apply workflow used for persistent configuration changes. The question asks which explanation best matches the actual broken dependency.

Incorrect Answers

B:

The corresponding repair is: Extract a versioned reusable Terraform module with validated input variables for site-specific network parameters. That explanation applies to a different technical failure. Here the controlling principle is: The corresponding repair is: Select the tool and deployment pattern by risk, state persistence, change scope, rollback capability and approval obligations.

C:

The offered statement is true for this other mechanism: The corresponding repair is: Read intended configuration and runtime operational YANG data from their appropriate modeled resources, and write only configurable leaves. It does not explain the facts described here, which follow from: The corresponding repair is: Select the tool and deployment pattern by risk, state persistence, change scope, rollback capability and approval obligations.

D:

This principle describes a separate cause: The corresponding repair is: Replace non-idempotent shell appends with the platform-aware Ansible network configuration module and explicit desired interface state. The reported condition is governed by: The corresponding repair is: Select the tool and deployment pattern by risk, state persistence, change scope, rollback capability and approval obligations. The two mechanisms should not be conflated.

Question 494

At fictional Amberlight Energy, an acceptance review of an AI-assisted code review finds: An incident chatbot interprets a troubleshooting question as authority to restart an OSPF process on production edge routers. Which targeted correction best addresses the network automation failure?

  1. Define typed, validated tool arguments with explicit device allowlists and reject invalid or unauthorized inputs at the server boundary.
  2. Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation.
  3. Return structured success and error states with source timestamps and bounded API timeouts rather than silently substituting empty values.
  4. Evaluate the agent on a representative labeled test set with difficult, adversarial and negative cases, measuring factual error and unsafe-action rates.

Explanation

Correct Answer

B:

Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. This addresses the problem because agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. Confirm success using: A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms.

Incorrect Answers

A:

typed MCP tool schemas improve interoperability but server-side validation remains necessary even for a well-formed client call. That explanation applies to a different technical failure. Here the controlling principle is: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated.

C:

This principle describes a separate cause: tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. The reported condition is governed by: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. The two mechanisms should not be conflated.

D:

high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. That explanation applies to a different technical failure. Here the controlling principle is: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated.

Question 495

At fictional Marble Bay Software, an acceptance review of an automation service security audit finds: A device automation web endpoint concatenates user-supplied interface names into a shell command executed on the runner. What should the team add to its network automation operating process?

  1. Maintain tested parser/device version matrices and fail clearly when collection contracts change.
  2. Add typed error classification and separate authorization and authentication runbooks to API clients.
  3. Version baseline snapshots and specify per-feature allowed changes rather than requiring exact equality for volatile counters.
  4. Add injection regression tests and prohibit unmanaged shell execution in network automation services.

Explanation

Correct Answer

D:

Add injection regression tests and prohibit unmanaged shell execution in network automation services. This prevents the recurrence because untrusted input can cross command or configuration boundaries; validation and safe APIs prevent injection into privileged operations. A matching immediate correction is: Use typed input schema validation, approved resource allowlists and safe library argument interfaces rather than shell concatenation. Applied here, the relevant facts are: A device automation web endpoint concatenates user-supplied interface names into a shell command executed on the runner. The question asks what should the team add to its network automation operating process.

Incorrect Answers

A:

The offered action targets: model-driven test tools depend on supported device schemas and parser versions; missing parsed data is not proof of missing network function. The case facts point instead to: untrusted input can cross command or configuration boundaries; validation and safe APIs prevent injection into privileged operations. Implement “Add injection regression tests and prohibit unmanaged shell execution in network automation services” to address that specific condition.

B:

This alternative is a response to another technical cause: hTTP 401 generally signals missing or invalid authentication while 403 indicates access denied to an authenticated request. The reported case instead depends on: untrusted input can cross command or configuration boundaries; validation and safe APIs prevent injection into privileged operations. The action that directly addresses it is “Add injection regression tests and prohibit unmanaged shell execution in network automation services”.

C:

The proposed action belongs to the mechanism: pyATS can learn and compare device operational state to detect unintended network effects beyond text config diffs. Here the stated fault follows a different mechanism: untrusted input can cross command or configuration boundaries; validation and safe APIs prevent injection into privileged operations. The corrective choice is “Add injection regression tests and prohibit unmanaged shell execution in network automation services”.

Question 496

While investigating a source-of-truth integration for fictional Willowfield Hospitality, the team reports: A GitLab CI build succeeds on a developer laptop but fails in the runner because the automation package was never declared as a dependency. What should the engineer change first to resolve this specific issue?

  1. Produce a reproducible versioned build artifact in the build stage and use that immutable artifact for subsequent validation and deployment.
  2. Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout.
  3. Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment.
  4. Make deployment depend on successful prevalidation jobs for schema, policy, simulation and approval checks.

Explanation

Correct Answer

C:

Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment. This addresses the problem because cI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. Confirm success using: A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state. Applied here, the relevant facts are: A GitLab CI build succeeds on a developer laptop but fails in the runner because the automation package was never declared as a dependency. The question asks what should the engineer change first to resolve this specific issue.

Incorrect Answers

A:

This alternative is a response to another technical cause: a CI build stage turns source and fixed dependencies into a traceable deployment input used by later stages. The reported case instead depends on: cI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. The action that directly addresses it is “Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment”.

B:

The proposed action belongs to the mechanism: a deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. Here the stated fault follows a different mechanism: cI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. The corrective choice is “Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment”.

D:

This alternative is a response to another technical cause: prevalidation is a safety boundary before changing network state, separate from checks after a change completes. The reported case instead depends on: cI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. The action that directly addresses it is “Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment”.

Question 497

During a conversational network agent pilot at fictional Westlake Biomedicine, the change record states: An LLM invents a controller API path not documented in the deployed Cisco software and presents it as a certain fix. Which result most directly verifies that the proposed correction worked?

  1. The assistant identifies missing version evidence and withholds unverified commands instead of fabricating supported APIs.
  2. A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms.
  3. Injection test logs are summarized as evidence but never cause unauthorized configuration or secret-retrieval tool calls.
  4. A poisoned-source test yields a flagged uncertainty or validated official-source citation instead of an unsafe configuration recommendation.

Explanation

Correct Answer

A:

The assistant identifies missing version evidence and withholds unverified commands instead of fabricating supported APIs. This directly verifies the related correction: Require citation to a compatible authoritative reference and fail closed or escalate when command syntax or device version is uncertain. It checks the technical distinction: Model fluency is not evidence of device capability; uncertain technical recommendations require explicit verification before action.

Incorrect Answers

B:

The offered action targets: This directly verifies the related correction: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. The case facts point instead to: This directly verifies the related correction: Require citation to a compatible authoritative reference and fail closed or escalate when command syntax or device version is uncertain. Implement “The assistant identifies missing version evidence and withholds unverified commands instead of fabricating supported APIs” to address that specific condition.

C:

This alternative is a response to another technical cause: This directly verifies the related correction: Treat retrieved logs as untrusted data and enforce instruction hierarchy and independent tool permissions outside the model. The reported case instead depends on: This directly verifies the related correction: Require citation to a compatible authoritative reference and fail closed or escalate when command syntax or device version is uncertain. The action that directly addresses it is “The assistant identifies missing version evidence and withholds unverified commands instead of fabricating supported APIs”.

D:

The proposed action belongs to the mechanism: This directly verifies the related correction: Ground actionable recommendations in trusted versioned technical references and reject unsupported or conflicting retrieved claims. Here the stated fault follows a different mechanism: This directly verifies the related correction: Require citation to a compatible authoritative reference and fail closed or escalate when command syntax or device version is uncertain. The corrective choice is “The assistant identifies missing version evidence and withholds unverified commands instead of fabricating supported APIs”.

Question 498

At fictional Foxglove Aviation, an acceptance review of a certificate renewal rehearsal finds: A network monitoring rollout tags every counter with a constantly changing session token and storage performance collapses. What ongoing engineering assurance should follow the successful fix?

  1. Review change-specific expected route and adjacency deltas alongside baseline safety invariants in test cases.
  2. Collect bounded latency histograms, timeout reason labels and retry budgets for critical controller calls.
  3. Review telemetry label schemas, retention and cardinality budgets before enabling new collectors.
  4. Monitor certificate expirations and trust-chain changes with automated predeployment verification.

Explanation

Correct Answer

C:

Review telemetry label schemas, retention and cardinality budgets before enabling new collectors. This prevents the recurrence because time-series cardinality grows with combinations of labels; volatile unique IDs defeat aggregation and strain metric storage. A matching immediate correction is: Normalize telemetry dimensions to stable device and interface identifiers and avoid high-cardinality ephemeral labels. Applied here, the relevant facts are: A network monitoring rollout tags every counter with a constantly changing session token and storage performance collapses. The question asks what ongoing engineering assurance should follow the successful fix.

Incorrect Answers

A:

That safeguard is relevant to: postdeployment validation compares expected invariants and intended deltas, not every volatile or purposely changed value. The current failure instead follows from: time-series cardinality grows with combinations of labels; volatile unique IDs defeat aggregation and strain metric storage. Its prevention measure is “Review telemetry label schemas, retention and cardinality budgets before enabling new collectors”.

B:

This routine guards against another mechanism: latency and retry failures require timing context and per-attempt telemetry; an aggregate success percentage is insufficient. The recurring exposure in the current case is: time-series cardinality grows with combinations of labels; volatile unique IDs defeat aggregation and strain metric storage. The matching preventive check is “Review telemetry label schemas, retention and cardinality budgets before enabling new collectors”.

D:

That safeguard is relevant to: tLS server validation depends on trusted issuer chains and hostname identity; bypassing verification defeats server authentication. The current failure instead follows from: time-series cardinality grows with combinations of labels; volatile unique IDs defeat aggregation and strain metric storage. Its prevention measure is “Review telemetry label schemas, retention and cardinality budgets before enabling new collectors”.

Question 499

During a production change postcheck at fictional Marble Bay Software, the change record states: A gNMI telemetry collector cannot process bursty device messages and drops interface counters without reporting any gaps. What mechanism is most directly responsible for these symptoms?

  1. Distributed troubleshooting requires durable transport and consistent time context; ephemeral local logs can be lost or misordered.
  2. Streaming telemetry pipelines have collectors, transports, ingestion queues and storage capacity that can fail independently of device subscriptions.
  3. Hardcoded secrets create persistent exposure and excessive authority; runtime secret retrieval limits both lifetime and access scope.
  4. Webhooks are asynchronous network events; integrity, delivery acknowledgements and idempotent processing are separate reliability needs.

Explanation

Correct Answer

B:

Streaming telemetry pipelines have collectors, transports, ingestion queues and storage capacity that can fail independently of device subscriptions. The corresponding repair is: Introduce bounded buffering, backpressure monitoring and horizontal ingestion capacity with visible gap or loss reporting. Acceptance should show: Load tests show telemetry consumers keeping up with target throughput or explicitly signaling lost samples without presenting false continuity. Applied here, the relevant facts are: A gNMI telemetry collector cannot process bursty device messages and drops interface counters without reporting any gaps. The question asks what mechanism is most directly responsible for these symptoms.

Incorrect Answers

A:

This principle describes a separate cause: The corresponding repair is: Synchronize clocks and ship logs to a durable centralized destination with appropriate secure, reliable transport. The reported condition is governed by: The corresponding repair is: Introduce bounded buffering, backpressure monitoring and horizontal ingestion capacity with visible gap or loss reporting. The two mechanisms should not be conflated.

C:

The offered statement is true for this other mechanism: The corresponding repair is: Inject short-lived narrowly scoped secrets at runtime from a managed vault and mask them in logs and build artifacts. It does not explain the facts described here, which follow from: The corresponding repair is: Introduce bounded buffering, backpressure monitoring and horizontal ingestion capacity with visible gap or loss reporting.

D:

This principle describes a separate cause: The corresponding repair is: Authenticate webhook delivery and implement bounded retries with deduplication or delivery-state accounting. The reported condition is governed by: The corresponding repair is: Introduce bounded buffering, backpressure monitoring and horizontal ingestion capacity with visible gap or loss reporting. The two mechanisms should not be conflated.

Question 500

While investigating a FastMCP security review for fictional Elmcrest Theatre, the team reports: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. Which targeted correction best addresses the network automation failure?

  1. Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules.
  2. Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints.
  3. Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation.
  4. Review generated code provenance, included dependencies and organizational IP policy before accepting it into the repository.

Explanation

Correct Answer

B:

Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints. This addresses the problem because tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. Confirm success using: A simulated malicious tool response cannot trigger outbound requests to unapproved destinations or reveal hidden credentials. Applied here, the relevant facts are: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. The question asks which targeted correction best addresses the network automation failure.

Incorrect Answers

A:

The proposed action belongs to the mechanism: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.

C:

This alternative is a response to another technical cause: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. The reported case instead depends on: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The action that directly addresses it is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.

D:

The proposed action belongs to the mechanism: aI-generated code may embed third-party patterns or dependencies; tool output is not automatically cleared of licensing obligations. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.