Practice Exams:

Palo Alto Networks NetSec-Architect Dumps

EXAM: Palo Alto Networks Network Security Architect
Verified By Experts
Premium Bundle
$49.99

NetSec-Architect Premium ETE File


  • Premium File 45 Questions & Answers
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Palo Alto Networks NetSec-Architect Practice Test 8

Question 70

A software company must identify after-hours firewall rule drift affecting protected production APIs. Suppose the team implements this unsafe shortcut: Perform a one-time vulnerability scan and declare every existing trusted session safe indefinitely. Which concrete security or availability failure is most likely?

  1. Newly connected unmanaged devices gain business application access before ownership and risk are established
  2. Later compromised identities and policy drift may remain unobserved between periodic audits
  3. Compromised user endpoints can pivot to sensitive data hosts with no application-specific control
  4. A compromised workload may traverse application tiers unnoticed even though the perimeter is correctly filtered

Answer & Explanations

Correct answer: B

A.

That outcome arises from a different failure: Newly connected unmanaged devices gain business application access before ownership and risk are established. The related situation is: A hospital introduces unfamiliar diagnostic carts to a segment where patient records must remain restricted. In the present case: A software company must identify after-hours firewall rule drift affecting protected production APIs. The more pertinent consequence is: Later compromised identities and policy drift may remain unobserved between periodic audits. Prevent it by: Design persistent analytics of authentication, endpoint posture, security policy changes and allowed-session threat logs with investigation ownership.

B.

Later compromised identities and policy drift may remain unobserved between periodic audits. This follows when the team chooses the shortcut: Perform a one-time vulnerability scan and declare every existing trusted session safe indefinitely. The situation is: A software company must identify after-hours firewall rule drift affecting protected production APIs. The correction is: Design persistent analytics of authentication, endpoint posture, security policy changes and allowed-session threat logs with investigation ownership. The observable test is: Correlate a revoked identity with changed session outcomes and confirm monitoring alerts route to an accountable incident team.

C.

That outcome arises from a different failure: Compromised user endpoints can pivot to sensitive data hosts with no application-specific control. The related situation is: A university must isolate a student lab from its registrar database while allowing approved portal queries. In the present case: A software company must identify after-hours firewall rule drift affecting protected production APIs. The more pertinent consequence is: Later compromised identities and policy drift may remain unobserved between periodic audits. Prevent it by: Design persistent analytics of authentication, endpoint posture, security policy changes and allowed-session threat logs with investigation ownership.

D.

That outcome arises from a different failure: A compromised workload may traverse application tiers unnoticed even though the perimeter is correctly filtered. The related situation is: A payment processor isolates API workers from adjacent databases inside a shared virtual cluster. In the present case: A software company must identify after-hours firewall rule drift affecting protected production APIs. The more pertinent consequence is: Later compromised identities and policy drift may remain unobserved between periodic audits. Prevent it by: Design persistent analytics of authentication, endpoint posture, security policy changes and allowed-session threat logs with investigation ownership.

Question 71

A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Which network security architecture decision best satisfies the business and technical requirements?

  1. Apply sanctioned AI-application policy plus Enterprise DLP inspection to controlled prompt submissions and model outputs
  2. Classify discovered AI SaaS applications, sanction reviewed business tools and apply measured access and data actions to unknown tools
  3. Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion
  4. Document AI architecture controls against applicable privacy, data residency and industry obligations before allowing regulated data flows

Answer & Explanations

Correct answer: C

A.

That choice fits Prevent sensitive data leakage through AI service content, illustrated by: A hospital allows a sanctioned AI summarizer but must restrict transmission of patient identifiers. Here the material condition is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The controlling reason is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. Apply: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

This action would be appropriate when: A consulting firm approves one AI meeting summarizer but employees start using unreviewed clone apps. It follows from: App-ID Cloud Engine and web controls identify applications, while DLP and threat controls limit data movement independent of a simple allowlist. The current case instead states: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The response it requires is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

C.

The best-matched action is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. The actual need is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Here Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. Check the result by: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

D.

That choice fits Map AI use to jurisdiction and industry governance requirements, illustrated by: A European healthcare research team assesses patient-data transfer to a third-country AI inference service. Here the material condition is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The controlling reason is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. Apply: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

Question 72

A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. An architect proposes: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Which technical explanation most directly supports that architecture?

  1. Framework obligations govern storage, purpose, retention and access; a technically encrypted model API may still violate contractual or statutory restrictions
  2. DLP exceptions permit defined business processing without broad disablement and maintain audit evidence for any sensitive content handling
  3. Adversarial tests simulate prompt injection, data extraction or agent misuse, revealing behavior that static model file scanning cannot validate
  4. Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment

Answer & Explanations

Correct answer: D

A.

This offered mechanism is relevant to Map AI use to jurisdiction and industry governance requirements: Framework obligations govern storage, purpose, retention and access; a technically encrypted model API may still violate contractual or statutory restrictions. It explains the different task: Document AI architecture controls against applicable privacy, data residency and industry obligations before allowing regulated data flows. The current condition is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

B.

This offered mechanism is relevant to Enforce user-level AI application governance with DLP exceptions: DLP exceptions permit defined business processing without broad disablement and maintain audit evidence for any sensitive content handling. It explains the different task: Use narrowly scoped policy exceptions for specific sanctioned AI workflows with documentable owners and expiry. The current condition is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

C.

This offered mechanism is relevant to Use adversarial AI red teaming before business release: Adversarial tests simulate prompt injection, data extraction or agent misuse, revealing behavior that static model file scanning cannot validate. It explains the different task: Schedule controlled Prisma AIRS red-team tests against target model prompts, tools, guardrails and permitted data access. The current condition is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

D.

Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. This causal relationship matters in: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. It supports the action: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Test the result by: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

Question 73

A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Which observable test most directly confirms the proposed architecture meets its intended security and availability outcomes?

  1. Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked
  2. Record reproducible benign attack simulations and confirm remediation changes prevent the same misuse before rollout
  3. Send safe test prompts through the actual inference path and verify policy actions plus traceable incident records
  4. Exercise a harmless unauthorized tool request and verify denial, auditing and business approval for a permitted action

Answer & Explanations

Correct answer: A

A.

Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked. This evidence tests whether the response, Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion, addresses the specific condition: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The mechanism being tested is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

B.

That test checks a different intervention: Schedule controlled Prisma AIRS red-team tests against target model prompts, tools, guardrails and permitted data access. Its applicable setting is: A bank prepares a customer assistant that may call account-lookup tools and needs prompt-injection testing. It would not directly establish the result required by: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

C.

That test checks a different intervention: Place approved Prisma AIRS runtime inspection on the supported inference path with appropriate detection and enforcement policy. Its applicable setting is: A travel platform exposes a model-backed booking assistant and needs continuous misuse controls in production. It would not directly establish the result required by: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

D.

That test checks a different intervention: Limit agent tool access using service identities, approval boundaries and monitored AI runtime policy instead of granting broad automation credentials. Its applicable setting is: A procurement assistant can create drafts but must not approve payments without a human control gate. It would not directly establish the result required by: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

Question 74

A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Which tempting shortcut would most directly undermine the required architecture control?

  1. Give every AI agent unrestricted administrator tokens because tool calling should operate without interruptions
  2. Depend only on a production web firewall to inspect a model checkpoint before it is executed
  3. Deploy only a browser extension while allowing unrestricted backend AI API traffic inside a shared VPC
  4. Pick a security appliance solely by throughput rating without mapping bypass or encrypted application routes

Answer & Explanations

Correct answer: B

A.

This shortcut would undermine Safeguard AI agents that act on enterprise tools for: A procurement assistant can create drafts but must not approve payments without a human control gate. The present situation is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

The unsafe shortcut is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. It conflicts with the necessary control for: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The plausible consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Instead: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Confirm with: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

C.

This shortcut would undermine Choose an AI security architecture for internal inference for: A manufacturer hosts proprietary vision models on Kubernetes and needs guarded API-to-model communications. The present situation is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

D.

This shortcut would undermine Select AI runtime control form factor from traffic topology for: A SaaS builder runs public inference APIs and private batch calls that require different inspection insertion points. The present situation is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

Question 75

A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. Suppose the team implements this unsafe shortcut: Depend only on a production web firewall to inspect a model checkpoint before it is executed. Which concrete security or availability failure is most likely?

  1. Critical inference APIs may evade controls even though security dashboards show some inspected traffic
  2. Regulated records can leave the organization through authorized model prompts or generated output
  3. A malicious model artifact can enter production despite apparently safe HTTP inference traffic
  4. Unreviewed AI services can collect enterprise content outside approved retention and compliance controls

Answer & Explanations

Correct answer: C

A.

That outcome arises from a different failure: Critical inference APIs may evade controls even though security dashboards show some inspected traffic. The related situation is: A SaaS builder runs public inference APIs and private batch calls that require different inspection insertion points. In the present case: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

That outcome arises from a different failure: Regulated records can leave the organization through authorized model prompts or generated output. The related situation is: A hospital allows a sanctioned AI summarizer but must restrict transmission of patient identifiers. In the present case: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

C.

A malicious model artifact can enter production despite apparently safe HTTP inference traffic. This follows when the team chooses the shortcut: Depend only on a production web firewall to inspect a model checkpoint before it is executed. The situation is: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The correction is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. The observable test is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

D.

That outcome arises from a different failure: Unreviewed AI services can collect enterprise content outside approved retention and compliance controls. The related situation is: A consulting firm approves one AI meeting summarizer but employees start using unreviewed clone apps. In the present case: A legal analytics startup downloads an external language-model checkpoint before hosting it on an internal inference platform. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

Question 76

An insurer evaluates supplier-provided vision models before approving a new automated claims service. Which network security architecture decision best satisfies the business and technical requirements?

  1. Place approved Prisma AIRS runtime inspection on the supported inference path with appropriate detection and enforcement policy
  2. Limit agent tool access using service identities, approval boundaries and monitored AI runtime policy instead of granting broad automation credentials
  3. Place model-serving components behind policy enforcement with segmented service paths and dedicated runtime inspection appropriate to the workload
  4. Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion

Answer & Explanations

Correct answer: D

A.

The alternative action addresses a different condition: A travel platform exposes a model-backed booking assistant and needs continuous misuse controls in production. Its reasoning is: Runtime protection observes request and response behavior during serving, distinct from predeployment red teaming and static model scanning. Here the case is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The more direct response is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

The alternative action addresses a different condition: A procurement assistant can create drafts but must not approve payments without a human control gate. Its reasoning is: Agent actions can change data or invoke APIs; least-privilege tool authorization and runtime oversight limit harm from prompt injection or tool misselection. Here the case is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The more direct response is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

C.

That choice fits Choose an AI security architecture for internal inference, illustrated by: A manufacturer hosts proprietary vision models on Kubernetes and needs guarded API-to-model communications. Here the material condition is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The controlling reason is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. Apply: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

D.

Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. The case describes: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The reason this response fits is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. Confirm it with: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

Question 77

An insurer evaluates supplier-provided vision models before approving a new automated claims service. An architect proposes: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Which technical explanation most directly supports that architecture?

  1. Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment
  2. An internal inference cluster needs network identity, application controls and runtime checks; an external browser-only control does not protect its server-side model calls
  3. The form factor determines observable prompts and responses, latency and policy consistency, so one deployment pattern cannot protect every AI traffic path
  4. AI requests can embed regulated data inside otherwise legitimate text, making identity-only app approval insufficient for content security

Answer & Explanations

Correct answer: A

A.

Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment. This causal relationship matters in: An insurer evaluates supplier-provided vision models before approving a new automated claims service. It supports the action: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Test the result by: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

B.

This offered mechanism is relevant to Choose an AI security architecture for internal inference: An internal inference cluster needs network identity, application controls and runtime checks; an external browser-only control does not protect its server-side model calls. It explains the different task: Place model-serving components behind policy enforcement with segmented service paths and dedicated runtime inspection appropriate to the workload. The current condition is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

C.

This offered mechanism is relevant to Select AI runtime control form factor from traffic topology: The form factor determines observable prompts and responses, latency and policy consistency, so one deployment pattern cannot protect every AI traffic path. It explains the different task: Choose supported inline network or workload-integrated AI inspection form factor based on where model inputs and outputs actually traverse. The current condition is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

D.

This offered mechanism is relevant to Prevent sensitive data leakage through AI service content: AI requests can embed regulated data inside otherwise legitimate text, making identity-only app approval insufficient for content security. It explains the different task: Apply sanctioned AI-application policy plus Enterprise DLP inspection to controlled prompt submissions and model outputs. The current condition is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. For that condition the operative principle is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

Question 78

An insurer evaluates supplier-provided vision models before approving a new automated claims service. Which observable test most directly confirms the proposed architecture meets its intended security and availability outcomes?

  1. Submit a synthetic sensitive identifier to a permitted AI service and verify blocking or alerting as defined
  2. Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked
  3. Verify allowed sanctioned AI use and blocked unapproved AI endpoints from separate test user groups
  4. Trace representative data fields through hosting, processing, logging and deletion controls under approved compliance criteria

Answer & Explanations

Correct answer: B

A.

That test checks a different intervention: Apply sanctioned AI-application policy plus Enterprise DLP inspection to controlled prompt submissions and model outputs. Its applicable setting is: A hospital allows a sanctioned AI summarizer but must restrict transmission of patient identifiers. It would not directly establish the result required by: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

B.

Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked. This evidence tests whether the response, Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion, addresses the specific condition: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The mechanism being tested is: Scanning evaluates model files and supply-chain risk at a defined lifecycle gate, distinct from monitoring inference prompts and outputs after deployment.

C.

That test checks a different intervention: Classify discovered AI SaaS applications, sanction reviewed business tools and apply measured access and data actions to unknown tools. Its applicable setting is: A consulting firm approves one AI meeting summarizer but employees start using unreviewed clone apps. It would not directly establish the result required by: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

D.

That test checks a different intervention: Document AI architecture controls against applicable privacy, data residency and industry obligations before allowing regulated data flows. Its applicable setting is: A European healthcare research team assesses patient-data transfer to a third-country AI inference service. It would not directly establish the result required by: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The specific acceptance check here is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

Question 79

An insurer evaluates supplier-provided vision models before approving a new automated claims service. Which tempting shortcut would most directly undermine the required architecture control?

  1. Assume using a vendor with an encryption checkbox automatically satisfies every regional privacy or sector requirement
  2. Create a permanent all-users DLP bypass for AI uploads after one legitimate analyst request fails
  3. Depend only on a production web firewall to inspect a model checkpoint before it is executed
  4. Conclude a generative AI assistant is safe merely because its model file has no known malware signature

Answer & Explanations

Correct answer: C

A.

This shortcut would undermine Map AI use to jurisdiction and industry governance requirements for: A European healthcare research team assesses patient-data transfer to a third-country AI inference service. The present situation is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

This shortcut would undermine Enforce user-level AI application governance with DLP exceptions for: A research analyst needs to submit public abstracts while private manuscripts remain blocked by AI upload DLP. The present situation is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

C.

The unsafe shortcut is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. It conflicts with the necessary control for: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The plausible consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Instead: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. Confirm with: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

D.

This shortcut would undermine Use adversarial AI red teaming before business release for: A bank prepares a customer assistant that may call account-lookup tools and needs prompt-injection testing. The present situation is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. Here the unsafe move is: Depend only on a production web firewall to inspect a model checkpoint before it is executed. That could cause: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Use: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

Question 80

An insurer evaluates supplier-provided vision models before approving a new automated claims service. Suppose the team implements this unsafe shortcut: Depend only on a production web firewall to inspect a model checkpoint before it is executed. Which concrete security or availability failure is most likely?

  1. Unsafe tool calling or sensitive-data disclosure paths can reach users even with a clean artifact scan
  2. Adversarial prompts or sensitive output can bypass protections that are absent during live inference
  3. A manipulated prompt can trigger privileged writes or extraction beyond the intended business workflow
  4. A malicious model artifact can enter production despite apparently safe HTTP inference traffic

Answer & Explanations

Correct answer: D

A.

That outcome arises from a different failure: Unsafe tool calling or sensitive-data disclosure paths can reach users even with a clean artifact scan. The related situation is: A bank prepares a customer assistant that may call account-lookup tools and needs prompt-injection testing. In the present case: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

B.

That outcome arises from a different failure: Adversarial prompts or sensitive output can bypass protections that are absent during live inference. The related situation is: A travel platform exposes a model-backed booking assistant and needs continuous misuse controls in production. In the present case: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

C.

That outcome arises from a different failure: A manipulated prompt can trigger privileged writes or extraction beyond the intended business workflow. The related situation is: A procurement assistant can create drafts but must not approve payments without a human control gate. In the present case: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The more pertinent consequence is: A malicious model artifact can enter production despite apparently safe HTTP inference traffic. Prevent it by: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion.

D.

A malicious model artifact can enter production despite apparently safe HTTP inference traffic. This follows when the team chooses the shortcut: Depend only on a production web firewall to inspect a model checkpoint before it is executed. The situation is: An insurer evaluates supplier-provided vision models before approving a new automated claims service. The correction is: Include Prisma AIRS model scanning in the release pipeline and require assessment of untrusted model artifacts before production promotion. The observable test is: Compare an approved model scan report with deployment admission and verify an intentionally rejected artifact is blocked.

Palo Alto Networks Network Security Architect (NetSec-Architect) Practice Tests