Basic Concept: AI-specific threat modeling requires consulting resources that catalogue adversarial attacks specifically developed for AI and ML systems. General cybersecurity frameworks may miss AI-unique attack vectors such as model inversion, data poisoning, and adversarial examples. CompTIA SecAI+ Study Guide identifies MITRE ATLAS as the authoritative source for AI system TTPs.

Why D is Correct: MITRE ATLAS provides a comprehensive, curated knowledge base of adversarial tactics, techniques, and procedures specifically targeting AI and ML systems, derived from real-world attack case studies. Analyzing ATLAS enables the architect to enumerate realistic AI-specific attacks applicable to the system being threat-modeled, which directly answers the question of which attacks can be performed.

Why A is Wrong: Using an LLM to map attack paths introduces uncertainty and potential hallucination risk. LLMs may generate plausible-sounding but inaccurate attack paths and cannot guarantee comprehensive coverage of AI-specific attack techniques.

Why B is Wrong: Quantifying risk of known vulnerabilities is a risk assessment step that occurs after identifying which attacks are possible. The architect must first identify attack possibilities before quantifying their risk impact.

Why C is Wrong: OWASP Top 10 covers web application vulnerabilities and, in its LLM edition, certain LLM-specific risks. However, MITRE ATLAS provides a more comprehensive and structured catalog of AI and ML-specific adversarial TTPs for systematic threat modeling.