A security architect performs threat modeling of an AI system. The architect needs to determine which attacks can be performed against the system. Which of the following actions should the architect…

Question

Options

• ALeverage a large language model (LLM) to map likely attack paths based on the code base.
• BQuantify the risk of known vulnerabilities identified in the AI system.
• CIdentify trust boundaries and perform threat modeling with Open Worldwide Application Security
• DAnalyze MITRE Adversarial Threat Landscape for AI Systems (ATLAS) for tactics, techniques, and

Explanation

D is correct. MITRE ATLAS (Adversarial Threat Landscape for AI Systems) is a purpose-built knowledge base cataloging tactics, techniques, and procedures (TTPs) that adversaries use specifically against AI and machine learning systems - making it the ideal resource when the goal is identifying which attacks are possible against an AI system. It is the AI-specific analog to MITRE ATT&CK for enterprise environments.

Why the others are wrong:

• A is wrong because asking an LLM to map attack paths is informal and non-authoritative - it lacks the rigor and structure of an established security framework, and LLM outputs are not a recognized threat modeling methodology.
• B is wrong because quantifying risk belongs to the risk assessment phase, which happens after threats are identified - not during threat identification itself.
• C is wrong because while OWASP and trust boundary analysis are useful in general threat modeling (e.g., STRIDE methodology), they are not AI-specific and don't provide the catalogued adversarial techniques needed to answer "which attacks can be performed" on an AI system.

Memory tip: Think of MITRE ATLAS as "ATT&CK for AI" - just as ATT&CK lists adversary TTPs against enterprise systems, ATLAS lists TTPs specifically targeting AI/ML systems. When the exam says "AI system + determine attacks," go straight to ATLAS.

No community discussion yet for this question.