One of the most significant additions in the move from CASP+ to SecurityX is explicit AI and machine learning content in the exam objectives. The 2024 CAS-005 release added AI-aware design to Domain 2 (Architecture) and adversarial AI considerations to Domain 4 (Operations). The material has continued to evolve as CompTIA updates the question pool through 2026.
What the objectives actually cover
The CAS-005 objectives reference AI and ML in three specific contexts.
AI in security operations. Using ML models for anomaly detection, alert triage, and threat hunting. The exam tests whether you can evaluate when ML adds value to a SOC workflow and when it adds noise. The most common scenario: a question stem describes a SOC drowning in alerts and asks which intervention helps. The wrong answer is usually "deploy AI/ML"; the right answer involves detection engineering discipline first, with ML as a downstream optimization.
Securing AI-integrated applications. Threat models for systems that incorporate LLMs, traditional ML models, or AI APIs. The exam covers prompt injection, model supply chain risks, data leakage through model outputs, and the privacy implications of training data exposure. OWASP LLM Top 10 categories show up directly in question wording as of mid-2026.
Adversarial AI. How attackers use AI to scale their work (phishing generation, vulnerability discovery, deepfake-based social engineering) and how defenders should adapt. The exam tests detection patterns that account for AI-augmented attacker velocity rather than expecting you to recognize specific AI-generated artifacts.
The threat models that actually matter
If you have limited time to prepare for the AI content, two threat models account for most of the testable material.
The first is the LLM-integrated application threat model. An application that calls an LLM has new attack surfaces: prompt injection through user input, data leakage through over-permissioned context, supply chain risk from the model provider, and authorization bypass through cleverly constructed queries. The bootcamp Day 2 covers this end-to-end with a lab.
The second is the AI-augmented attacker threat model. Phishing volume up 4 to 10x according to multiple 2025 vendor reports. Social engineering enriched by public AI tools. Vulnerability research compressed from weeks to hours for routine bugs. The defender response is the same general posture as always, but tuned for higher attacker throughput: faster detection cycles, more aggressive credential hygiene, better deepfake-resistant verification flows for high-value transactions.
What the exam will not test
The exam will not ask you to name specific models, vendors, or AI products. It will not test deep ML knowledge such as model training mechanics, hyperparameter tuning, or specific algorithm internals. It treats AI/ML the way a senior security practitioner treats it: as a system with attack surface and a tool with limited applicability, not as a topic that demands a CS degree.
That framing matters. Candidates who study AI from a data-science angle often over-prepare on the wrong material. Study AI the way you would study any other piece of enterprise infrastructure: what are its inputs, what are its outputs, where does trust flow, and what breaks.
Where to focus prep time
Read NIST AI 100-1 (the AI Risk Management Framework), the OWASP Top 10 for LLMs, and at least one recent threat report describing AI-augmented attacks (CrowdStrike, Mandiant, and Microsoft all publish current data). Those three sources cover roughly 80 percent of the testable AI material on CAS-005 as of 2026.
Leave a Reply