About these questions
These five questions are original. We wrote them in the scenario style that the SecAI+ objectives suggest, with more weight on securing AI systems because that domain is 40% of the exam. They are not real exam questions, and the real exam will word things differently. The questions are all multiple-choice, so they do not show what CompTIA’s performance-based items look like.
How to use these
Try each question without looking at your notes, and pick an answer before you open the explanation. Then read the explanation even when you were right, because the reason the closest wrong option fails is where the learning is. Note any domain where you slipped and revisit it on the domains explained page.
Question 1 (Basic AI concepts related to cybersecurity)
A bank is preparing a fraud-detection dataset. Only 2% of the records are fraud cases. The file also holds many repeated rows and some records with corrupted date fields. The auditor requires that every training record be a genuine, recorded transaction, so no synthetic or generated samples may be added. No new data can be bought. Which TWO activities address the problems? Choose two.
- A. Data augmentation
- B. Data cleansing
- C. Quantization
- D. Data balancing
- E. Watermarking
Show the answer and explanation
Correct answers: B and D. Data cleansing removes the repeated rows and fixes or drops the corrupted records. Data balancing corrects the skew toward non-fraud cases, and it can be done with the existing records, for example by sampling or weighting them differently. Data augmentation is the tempting third choice because it also helps rare classes, but it creates synthetic samples, which the auditor has ruled out. Quantization shrinks a model and is unrelated to data quality, and watermarking marks content so it can be recognized later.
Question 2 (Securing AI systems)
A retail chatbot went live with guardrails that were tested and approved nine months ago. Last week the model provider moved the service to a newer version of the model. The application code and configuration did not change. What should the team do before continuing to rely on the guardrails?
- A. Re-run the existing set of adversarial and edge-case prompts against the new model version and review the results
- B. Increase the token limit and shorten the prompt template
- C. Run a software composition analysis on the application’s dependencies
- D. Note the provider’s release summary in the approval record and keep the original sign-off
Show the answer and explanation
Correct answer: A. The only thing that changed is the model underneath the guardrails, so the earlier test results no longer describe the system. Re-running the adversarial suite produces fresh evidence for the new version. The release summary in option D is the closest wrong answer: it is generic information about the provider’s change, not a test of this configuration. Option B changes settings without testing anything, and option C inspects dependencies that did not change.
Question 3 (Securing AI systems)
Users of a vendor-hosted support assistant get restricted answers by spreading a forbidden request across several messages and encoding parts of it. Each single message looks harmless in the logs. The team cannot retrain the model, and responses must still arrive within about two seconds. Which compensating control fits best?
- A. A prompt firewall at the gateway that checks inputs and outputs in the context of the whole session
- B. Fine-tuning the model to refuse split and encoded requests
- C. Per-user rate limits
- D. Encrypting the stored conversation logs
Show the answer and explanation
Correct answer: A. A prompt firewall sits at the gateway, so it works without changing the hosted model, and it can judge a message together with the earlier turns in the session. Fine-tuning is the closest wrong option, but the team cannot retrain a vendor-hosted model. Rate limiting restricts volume, and this attack needs only a handful of messages. Encrypting logs protects stored data but does nothing to stop the abusive requests.
Question 4 (AI-assisted security)
A team’s delivery pipeline runs unit tests, regression tests, code scanning, and software composition analysis. After the provider released a new model version, the quality of a threat-detection feature dropped in production, yet no pipeline stage failed. The application code had not changed. What should the team add to the pipeline?
- A. More regression tests over the application code
- B. Software composition analysis of the model server’s dependencies
- C. A model testing stage that scores each model version against a held-out benchmark with a pass threshold
- D. Extra unit tests for the code that calls the model
Show the answer and explanation
Correct answer: C. The model was the only moving part, so only a stage that measures the model can catch a drop in quality, and a held-out benchmark with a threshold gives a clear pass or fail. Regression and unit tests (options A and D) treat the code as the variable, and code did not change. Software composition analysis, the closest wrong option, reports on known vulnerabilities in dependencies, not on whether the model still performs well.
Question 5 (AI governance, risk, and compliance)
A lender must tell every rejected applicant, individually, which main factors led to the decision about that person. The lender already publishes a notice on its website saying that automated scoring is used. Which responsible-AI principle is still unmet?
- A. Fairness
- B. Transparency
- C. Explainability
- D. Reliability and safety
Show the answer and explanation
Correct answer: C. Giving the reasons behind one specific decision is explainability. Transparency, the closest wrong option, is the broader duty to disclose that AI is used and how it works, and the published notice already covers that. Fairness is about whether outcomes are equitable across groups, and reliability and safety are about dependable performance and safe failure. Neither is the gap described.
More guides for CY0-001
- CY0-001 Exam Format and Scoring
- CY0-001 Study Plan: 4 and 8 Weeks
- Is CY0-001 Hard? SecAI+ Difficulty
- CY0-001 Prerequisites and Skills Check
- CY0-001 Renewal and Retakes Explained
- Is CY0-001 Worth It? An Honest Look
- CY0-001 Cheat Sheet: Key Facts to Review
- SY0-801 vs CY0-001: Which Exam First?
Next step
If a question surprised you, the domains explained page shows where it fits. For more practice with an explanation on every answer option, try our CY0-001 practice tests.