OWASP LLM and ML Top 10 (2025) Mapping to AI Threat and Governance COASP Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   A financial institution uses an LLM to summarize news articles for traders. A trader notices that the summaries sometimes include fabricated quotes that appear credible. Which OWASP LLM Top 10 risk is this, and what governance control should be implemented?

   • AInsecure Output Handling; sanitize all summaries before display
   • BSensitive Information Disclosure; restrict access to the LLM
   • CPrompt Injection; add a system prompt that says 'do not hallucinate'
   • DHallucination; require the LLM to cite sources and implement fact-checking mechanisms
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   A security analyst is reviewing a machine learning model that classifies emails as spam or not spam. The model was trained on a dataset that included emails from a public source. The analyst discovers that an attacker could craft emails that are misclassified as legitimate by adding specific words. Which OWASP ML Top 10 risk is this, and what is a suitable mitigation?

   • AModel Inversion — restrict access to the model's confidence scores.
   • BData Poisoning — retrain the model on a clean dataset.
   • CAdversarial Examples — implement adversarial training and input sanitization.
   • DML Model Skew — monitor the model's performance over time.
   Reveal Solution Discussion Select an answer first
4. 38application · medium

   A company is deploying an LLM-based system that assists with legal document review. The system has access to confidential legal documents. The compliance team wants to ensure that the system does not inadvertently expose confidential information to unauthorized users. Which governance policy is most effective?

   • AUse a more powerful LLM that understands legal jargon
   • BLog all user interactions for audit purposes
   • CImplement role-based access control (RBAC) on the document repository and enforce output filtering
   • DRestrict the LLM to only answer questions about document summaries
   Reveal Solution Discussion Select an answer first
5. 39foundation · easy

   An organization is developing an AI governance policy. Which OWASP LLM Top 10 risk should they prioritize to address the potential for the model to be manipulated into performing unauthorized actions?

   • ALLM01: Prompt Injection
   • BLLM03: Supply Chain
   • CLLM05: Excessive Agency
   • DLLM02: Sensitive Information Disclosure
   Reveal Solution Discussion Select an answer first
6. 40foundation · easy

   A security researcher uses a series of queries to an image recognition API to reconstruct a training image of a person. Which OWASP ML Top 10 risk does this attack vector map to?

   • AML05: Model Inversion
   • BML07: Adversarial Perturbation
   • CML02: Data Poisoning
   • DML08: Model Theft
   Reveal Solution Discussion Select an answer first