CompTIA SecAI+ 2.0 Securing AI Systems Practice Questions

Question 1

Which of the following attacks involves an adversary manipulating the input data used to train a machine learning model to introduce a backdoor or bias?

Accepted Answer

Data poisoning

Answer

Model inversion

Answer

Membership inference

Answer

Sponge attack

Question 2

According to the OWASP Top 10 for LLM, which vulnerability occurs when an attacker manipulates the LLM through crafted inputs to bypass safety guardrails?

Accepted Answer

Prompt Injection

Answer

Model Theft

Answer

Supply Chain Vulnerabilities

Answer

Insecure Output Handling

Question 3

A security engineer is implementing a gateway control to prevent Denial of Service (DoS) attacks against an internal LLM API. Which control would be MOST effective?

Accepted Answer

Rate limiting

Answer

Prompt templates

Answer

Data lineage tracking

Answer

Model quantization

Question 4

Which attack type aims to reconstruct the training data or extract sensitive features of a specific individual from the model's output?

Accepted Answer

Model inversion

Answer

Model skewing

Answer

Prompt injection

Answer

Adversarial examples

Question 5

To secure an AI system's data at rest, which of the following is a standard cryptographic requirement?

Accepted Answer

AES-256 encryption

Answer

TLS 1.3

Answer

Data augmentation

Answer

Vector embedding

Question 6

An attacker is using an 'indirect prompt injection' technique. How is this attack typically delivered?

Accepted Answer

By embedding malicious instructions in a website or document that the LLM processes

Answer

By typing commands directly into the chatbot interface

Answer

By overloading the server with API requests

Answer

By stealing the model weights via a side-channel attack

Question 7

Which resource would be BEST for a security team to reference when looking for a knowledge base of adversary tactics and techniques specifically targeting AI systems?

Accepted Answer

MITRE ATLAS

Answer

MITRE ATT&CK

Answer

CVSS v3.1

Answer

NIST CSF

Question 8

A developer wants to ensure that the LLM only answers questions related to the company's HR policies and refuses all other topics. Which control should be implemented?

Accepted Answer

System prompt/System role

Answer

Data anonymization

Answer

Homomorphic encryption

Answer

Federated learning

Question 9

In the context of AI security, what does 'Jailbreaking' refer to?

Accepted Answer

Crafting prompts to bypass the model's ethical and safety content filters

Answer

Gaining root access to the server hosting the model

Answer

Decrypting the model weights

Answer

Escaping a containerized environment

Question 10

Which data processing technique involves replacing sensitive data identifiers with pseudonyms to protect privacy while maintaining analytical utility?

Accepted Answer

Data masking/Tokenization

Answer

Data poisoning

Answer

Data augmentation

Answer

Data pruning

Question 11

Which of the following represents a 'Membership Inference Attack'?

Accepted Answer

Determine if a specific data record was part of the model's training set

Answer

Stealing the entire model architecture

Answer

Flooding the model with requests to cause a crash

Answer

Injecting bias into the model

Question 12

What is the purpose of 'Red Teaming' in the context of LLM security?

Accepted Answer

To simulate adversarial attacks to identify vulnerabilities and harmful outputs

Answer

To optimize the model for speed

Answer

To clean the data before training

Answer

To monitor the cost of API tokens

Question 13

Which technique allows an attacker to manipulate an AI system by creating input that is visually indistinguishable to humans but misclassified by the computer?

Accepted Answer

Adversarial Example / Input Manipulation

Answer

SQL Injection

Answer

Cross-Site Scripting

Answer

Buffer Overflow

Question 14

Which of the following is a supply chain attack vector specific to AI?

Accepted Answer

Compromised Pre-trained Model from a public repository (e.g., Hugging Face)

Answer

Phishing email to the CEO

Answer

Weak Wi-Fi password

Answer

Unpatched Windows Server

Question 15

Which of the following is a method to mitigate 'Prompt Injection' attacks?

Accepted Answer

Using 'Sandboxing' or strictly separating user data from instructions

Answer

Increasing the temperature of the model

Answer

Giving the user root access

Answer

Disabling encryption

Question 16

What is a 'Sponge Attack' in the context of AI?

Accepted Answer

A Denial of Service (DoS) attack designed to maximize energy consumption and latency of the model

Answer

An attack that absorbs all the data from a database

Answer

Cleaning the data to remove impurities

Answer

A watermarking technique

Question 17

Which of the following is a method to detect 'Hallucinations' in LLM outputs?

Accepted Answer

Checking the response confidence level/score

Answer

Encrypting the prompt

Answer

Using a VPN

Answer

Increasing the model temperature

Question 18

What is the primary security concern regarding 'Excessive Agency' in AI agents?

Accepted Answer

The agent is granted permissions to perform actions (like sending emails or executing code) beyond what is necessary

Answer

The agent costs too much money

Answer

The agent responds too slowly

Answer

The agent has too many parameters

Question 19

An adversary uses a 'Transfer Learning Attack'. What are they doing?

Accepted Answer

Using a pre-trained model as a starting point to train a surrogate model that mimics a target victim model

Answer

Moving data from one server to another

Answer

Transferring money via AI banking

Answer

Uploading a virus to the cloud

Question 20

Which monitoring technique involves checking the output of an LLM for toxic, biased, or policy-violating content before showing it to the user?

Accepted Answer

Content moderation / Output filtering

Answer

Input validation

Answer

Rate limiting

Answer

Encryption

Question 21

Which organization publishes the 'AI Risk Repository' mentioned in the exam objectives?

Accepted Answer

MIT (Massachusetts Institute of Technology)

Answer

NASA

Answer

CompTIA

Answer

IEEE

Question 22

Which attack hides malicious instructions inside user-supplied content so an LLM ignores the system intent?

Accepted Answer

Prompt injection

Answer

SQL indexing

Answer

RAID rebuild

Answer

DNS recursion

Question 23

Which control helps prevent an AI assistant from returning secrets from connected tools?

Accepted Answer

Tool permission scoping and output filtering

Answer

Longer response temperature

Answer

Unrestricted file access

Answer

Disabling audit logs

Question 24

Which technique tests an AI system by intentionally crafting inputs to cause unsafe or incorrect behavior?

Accepted Answer

Adversarial testing

Answer

Disk formatting

Answer

MAC learning

Answer

Printer calibration

Question 25

Which AI security risk involves sensitive examples being exposed from training data?

Accepted Answer

Training data leakage

Answer

VLAN pruning

Answer

Drive fragmentation

Answer

NTP skew only

Question 26

Which control can reduce harmful LLM output before it reaches a user?

Accepted Answer

Content safety classifier or guardrail

Answer

Open relay

Answer

Unfiltered tool execution

Answer

Plaintext logging of secrets

Question 27

Which attack poisons examples so a model learns attacker-chosen behavior?

Accepted Answer

Data poisoning

Answer

Port mirroring

Answer

ARP aging

Answer

Normal validation

Question 28

Which security control limits damage if an AI tool is tricked into calling an external API?

Accepted Answer

Least-privilege API tokens

Answer

Permanent root tokens

Answer

Unbounded spending limits

Answer

Disabled monitoring

Question 29

Which monitoring signal may indicate an LLM application is under prompt injection attack?

Accepted Answer

Repeated requests to reveal hidden instructions or ignore prior rules

Answer

Normal TLS certificate renewal

Answer

Scheduled backups

Answer

Expected DNS cache hits

Question 30

Which control limits an LLM agent so it can read tickets but cannot delete them?

Accepted Answer

Tool-level least privilege

Answer

Higher temperature

Answer

Longer context only

Answer

Unfiltered chain output

Question 31

Which model attack attempts to infer whether a specific record was part of training data?

Accepted Answer

Membership inference

Answer

ARP spoofing

Answer

Port knocking

Answer

RAID striping

Question 32

Which risk can occur if retrieved documents contain attacker-written instructions?

Accepted Answer

Indirect prompt injection

Answer

CPU overheating

Answer

DHCP exhaustion only

Answer

Screen burn-in

Question 33

Which control reduces the chance an AI assistant exposes another tenant data?

Accepted Answer

Tenant isolation controls

Answer

Shared global memory

Answer

Public logs

Answer

Single API key for all tenants

Question 34

Which term describes intentionally crafted input that causes a model to misclassify?

Accepted Answer

Adversarial example

Answer

DNS record

Answer

Patch cable

Answer

Storage snapshot

Question 35

Which AI system risk is reduced by human approval before high-impact actions execute?

Accepted Answer

Autonomous action causing harm

Answer

Cable attenuation

Answer

Keyboard ghosting

Answer

NTP drift

Question 36

Which AI attack attempts to reconstruct parts of training data from model outputs?

Accepted Answer

Model inversion

Answer

MAC flooding

Answer

DNS tunneling only

Answer

VLAN hopping

Question 37

Which control helps stop an AI assistant from making unsupported factual claims?

Accepted Answer

Require citations to retrieved trusted sources

Answer

Increase randomness only

Answer

Disable input validation

Answer

Remove all context

CompTIA SecAI+ Practice Questions: 2.0 Securing AI Systems

Q1. Which of the following attacks involves an adversary manipulating the input data used to train a machine learning model to introduce a backdoor or bias?

Q2. According to the OWASP Top 10 for LLM, which vulnerability occurs when an attacker manipulates the LLM through crafted inputs to bypass safety guardrails?

Q3. A security engineer is implementing a gateway control to prevent Denial of Service (DoS) attacks against an internal LLM API. Which control would be MOST effective?

Q4. Which attack type aims to reconstruct the training data or extract sensitive features of a specific individual from the model's output?

Q5. To secure an AI system's data at rest, which of the following is a standard cryptographic requirement?

Q6. An attacker is using an 'indirect prompt injection' technique. How is this attack typically delivered?

Q7. Which resource would be BEST for a security team to reference when looking for a knowledge base of adversary tactics and techniques specifically targeting AI systems?

Q8. A developer wants to ensure that the LLM only answers questions related to the company's HR policies and refuses all other topics. Which control should be implemented?

Q9. In the context of AI security, what does 'Jailbreaking' refer to?

Q10. Which data processing technique involves replacing sensitive data identifiers with pseudonyms to protect privacy while maintaining analytical utility?

Q11. Which of the following represents a 'Membership Inference Attack'?

Q12. What is the purpose of 'Red Teaming' in the context of LLM security?

Q13. Which technique allows an attacker to manipulate an AI system by creating input that is visually indistinguishable to humans but misclassified by the computer?

Q14. Which of the following is a supply chain attack vector specific to AI?

Q15. Which of the following is a method to mitigate 'Prompt Injection' attacks?

Q16. What is a 'Sponge Attack' in the context of AI?

Q17. Which of the following is a method to detect 'Hallucinations' in LLM outputs?

Q18. What is the primary security concern regarding 'Excessive Agency' in AI agents?

Q19. An adversary uses a 'Transfer Learning Attack'. What are they doing?

Q20. Which monitoring technique involves checking the output of an LLM for toxic, biased, or policy-violating content before showing it to the user?

Q21. Which organization publishes the 'AI Risk Repository' mentioned in the exam objectives?

Q22. Which attack hides malicious instructions inside user-supplied content so an LLM ignores the system intent?

Q23. Which control helps prevent an AI assistant from returning secrets from connected tools?

Q24. Which technique tests an AI system by intentionally crafting inputs to cause unsafe or incorrect behavior?

Q25. Which AI security risk involves sensitive examples being exposed from training data?

Q26. Which control can reduce harmful LLM output before it reaches a user?

Q27. Which attack poisons examples so a model learns attacker-chosen behavior?

Q28. Which security control limits damage if an AI tool is tricked into calling an external API?

Q29. Which monitoring signal may indicate an LLM application is under prompt injection attack?

Q30. Which control limits an LLM agent so it can read tickets but cannot delete them?

Q31. Which model attack attempts to infer whether a specific record was part of training data?

Q32. Which risk can occur if retrieved documents contain attacker-written instructions?

Q33. Which control reduces the chance an AI assistant exposes another tenant data?

Q34. Which term describes intentionally crafted input that causes a model to misclassify?

Q35. Which AI system risk is reduced by human approval before high-impact actions execute?

Q36. Which AI attack attempts to reconstruct parts of training data from model outputs?

Q37. Which control helps stop an AI assistant from making unsupported factual claims?

More CompTIA SecAI+ practice topics