Free CY0-001 sample questions with answers
Try these before reading the answers. The mix reflects the real weighting, so two of the five come from securing AI systems.
Question 1. An attacker submits crafted records into the dataset used to retrain your fraud-detection model, so that transactions matching a pattern they control are later classified as legitimate. What is this attack called?
- A. Model extraction
- B. Data poisoning
- C. Prompt injection
- D. Denial of service
Show answer
Answer: B
Data poisoning manipulates training data so the resulting model behaves the way the attacker wants. Model extraction steals a copy of the model through repeated queries, prompt injection targets an instruction-following system at inference time, and denial of service targets availability rather than the model’s learned behaviour.
Question 2. A customer-facing assistant is built on a third-party model and must never disclose internal system instructions, even when users ask cleverly. Which control most directly addresses the risk?
- A. Encrypting the model weights at rest
- B. Network segmentation around the inference host
- C. Output filtering and guardrails on responses
- D. Increasing the model’s context window
Show answer
Answer: C
Output filtering and guardrails inspect what the system is about to return and block disclosure, which is the control that sits between the model and the user. Encryption at rest protects stored data, network segmentation limits lateral movement, and a longer context window is a capability setting with no security effect.
Question 3. Your SOC deploys an AI tool that triages alerts and closes those it judges benign. Six weeks later a genuine intrusion is found to have been auto-closed. Which weakness does this best illustrate?
- A. Over-reliance on AI decisions without human oversight
- B. Insufficient logging of network traffic
- C. Weak encryption of alert data in transit
- D. Poor network segmentation between SOC and production
Show answer
Answer: A
Over-reliance on automated judgement without human oversight allowed a false negative to become a missed intrusion. The problem is not that the tool was wrong once, but that nothing checked it. Insufficient logging, weak encryption and poor segmentation are real issues but none of them describe an alert being wrongly closed and never reviewed.
Question 4. Before deploying an AI system that influences hiring decisions, regulators expect you to document its purpose, the data it uses, its known limitations and who is accountable for it. What is this activity generally called?
- A. A penetration test
- B. A vulnerability scan
- C. A disaster recovery plan
- D. An AI impact assessment
Show answer
Answer: D
An AI impact assessment documents purpose, data, limitations, affected people and accountability before deployment, which is exactly the described set. A penetration test probes for exploitable weaknesses, a vulnerability scan finds known flaws in software, and a disaster recovery plan addresses restoring service after an outage.
Question 5. An application passes user-supplied text to a model that has permission to call internal tools. A user submits text containing instructions telling the model to ignore its rules and call an administrative tool. What is the underlying design flaw?
- A. The model is too large for the workload
- B. Untrusted input reaches a privileged system with no separation of data from instructions
- C. The API keys are not rotated frequently enough
- D. The application lacks a CAPTCHA on its input form
Show answer
Answer: B
Untrusted input is reaching an instruction-following system that holds privileges, with no separation between data and instructions and no constraint on which tools can be invoked. Choosing a smaller model, rotating keys or adding CAPTCHA would not change the fact that user text can direct a privileged system.
How did you do?
Notice what these questions did not require: no configuration syntax and no product names. SecAI+ tests whether you can identify the threat, name the control and reason about consequences. If the vocabulary in the explanations felt unfamiliar, start with week 1 of the study plan rather than diving into the securing domain.