OWASP LLM and ML Top 10 (2025) Mapping to AI Threat and Governance COASP Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A company is developing an LLM-powered code-generation tool for developers. The security team is concerned that the tool might generate code with vulnerabilities. Which OWASP LLM Top 10 risk is most relevant, and what governance control should be implemented?

   • ASensitive Information Disclosure; restrict access to the tool
   • BExcessive Agency; limit the tool's ability to execute code
   • CPrompt Injection; block all prompts that request code
   • DInsecure Output Handling; require the tool to run generated code in a sandbox
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   An attacker injects malicious samples into a training dataset used for a spam classifier, causing it to misclassify legitimate emails as spam. Which OWASP ML Top 10 risk does this attack vector map to?

   • AML05: Model Inversion
   • BML02: Data Poisoning
   • CML06: Membership Inference
   • DML07: Adversarial Perturbation
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A security team is evaluating a machine learning model that detects network intrusions. The team is concerned that an attacker could cause the model to misclassify malicious traffic as benign. Which OWASP ML Top 10 risk is this, and what offensive test should be performed?

   • AModel Inversion; attempt to extract training data from the model
   • BModel Evasion; craft adversarial network traffic to test if the model misses it
   • CData Poisoning; inject malicious traffic into the training set
   • DSupply Chain Attack; compromise the model's update server
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A security engineer is reviewing a customer-support chatbot that uses an LLM with retrieval-augmented generation (RAG) over internal knowledge bases. The chatbot is exposed to external users and occasionally outputs confidential information from the knowledge base that the user should not have access to. The engineer wants to map this issue to the OWASP LLM Top 10 and then implement a governance control. Which OWASP LLM risk best describes the issue, and which control should be prioritized?

   • APrompt Injection — add a system prompt that instructs the model to ignore malicious user inputs.
   • BSensitive Information Disclosure — implement strict access controls on the knowledge base and enforce output filtering.
   • CExcessive Agency — restrict the chatbot's ability to call external APIs and tools.
   • DInsecure Output Handling — sanitize all model outputs before displaying them to users.
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A red team is assessing an LLM-based system that automatically generates and sends emails to customers. The team discovers that an attacker can manipulate the LLM to send phishing emails to customers. Which OWASP LLM Top 10 risks are involved, and what is the most effective mitigation?

   • ASupply Chain Vulnerability and Overreliance; replace the LLM with a rule-based system
   • BHallucination and Model Evasion; add a content filter to block phishing keywords
   • CSensitive Information Disclosure and Insecure Output Handling; encrypt all emails
   • DPrompt Injection and Excessive Agency; implement human approval for all outgoing emails
   Reveal Solution Discussion Select an answer first