CompTIA SecAI+ study plan: 2-week and 4-week schedules

Here's a concrete, day-by-day plan for CompTIA SecAI+ (CY0-001): a 4-week track if you know security but AI is new to you, and a 2-week track if you already work in security and use LLM tools day to day. Follow either one and finish consistently scoring 80%+ on practice exams before test day.

SecAI+ is not an AI theory exam. It assumes you already think like a security professional and asks you to apply that to AI: how a prompt injection actually reaches a model, why a system prompt is not a security boundary, which control stops an over-privileged agent, and which framework a regulator expects. This plan is built around those judgment calls.

The exam you're planning for

Securing AI systems alone is 40% of the exam, so both plans give it the most days. The time limit is tight too: up to 60 questions in 60 minutes, including performance-based items. Pacing practice is part of the plan, not an afterthought.

Pick your track

The daily hour (both tracks)

Every study day follows the same shape:

• 20 minutes — learn: one topic from that day's list (the official objectives, OWASP and NIST docs, a course video, or your notes).
• 30 minutes — practice questions on that topic, plus a few review questions from earlier days.
• 10 minutes — review misses: read the explanation for every wrong answer and say out loud why the right control wins and the others don't.

The 4-week plan (new-to-AI track)

Week 1 — Basic AI concepts for security (17%)

• Days 1–2: AI vs ML vs deep learning; supervised, unsupervised and reinforcement learning, with a security example of each (phishing classifier, anomaly detection, adaptive defense). Transformers, LLMs vs small language models, GANs.
• Day 3: Training and refinement: validation sets, overfitting, epochs, fine-tuning, pruning and quantization. Data prep and integrity: cleansing, lineage, provenance, class balancing.
• Day 4: Prompting: system vs user prompts, zero/one/few-shot, prompt templates. Structured vs semi-structured vs unstructured data.
• Day 5: RAG, embeddings and vector databases (and why the vector store needs access control), watermarking, and the AI lifecycle from data collection to monitoring.
• Weekend: A 25-question quiz over the domain. Anything under ~70%, re-read that topic.

Week 2 — Securing AI systems, part 1: the attacks (40%)

• Day 1: The OWASP Top 10 for LLM Applications (2025). Learn every entry with one concrete example: prompt injection, sensitive information disclosure, supply chain, data and model poisoning, improper output handling, excessive agency, system prompt leakage, vector and embedding weaknesses, misinformation, unbounded consumption.
• Day 2: Direct vs indirect prompt injection and jailbreaks. Practice tracing how text in an email, web page or document reaches the model.
• Day 3: Attacks on the model itself: poisoning and backdoors, adversarial evasion, model inversion, membership inference, model extraction/theft. Malicious model files (unsafe pickle deserialization).
• Day 4: Threat-modeling resources: MITRE ATLAS (and how it maps to ATT&CK), the OWASP ML Security Top 10, and the MIT AI Risk Repository's entity/intent/timing taxonomy.
• Day 5: Mixed attack-identification drill: read a short incident, name the risk. Aim for instant recognition.
• Weekend: Domain quiz plus a re-drill of misses.

Week 3 — Securing AI systems, part 2: the controls

• Day 1: Model and gateway controls: guardrails on input and output, prompt templates, AI gateways and prompt firewalls, rate limits, token limits, quotas, modality restrictions.
• Day 2: Least privilege for models, data, agents, tools and APIs: scoped credentials, human approval for high-impact actions, sandboxed code execution, permission-aware retrieval.
• Day 3: Data protection: encryption in transit, at rest and in use (confidential computing), anonymization vs pseudonymization, classification, redaction and masking of PII in prompts and logs, data minimization.
• Day 4: Monitoring and supply chain: prompt/response logging, anomaly detection on queries, confidence scoring, cost monitoring, signed model artifacts, AI/ML bills of materials, model-file scanning.
• Day 5: Use the cheat sheet below until "attack → best control" is automatic.
• Weekend: Mixed quiz across weeks 1–3.

Week 4 — AI-assisted security, governance, and practice exams

• Day 1: AI-assisted security (24%): AI in the SOC (triage, summarization, UEBA, query generation), coding assistants and AI code scanning in CI/CD, MCP servers and agent tools, and the limits (hallucinated findings, data leakage into public tools, human validation).
• Day 2: Offensive AI: deepfake voice and video for BEC, AI-written phishing at scale, AI-driven bots and malware, and the defenses (out-of-band verification, behavior-based detection, layered bot management).
• Day 3: Governance, risk and compliance (19%): AI Center of Excellence, roles, acceptable use and shadow AI, responsible-AI principles, NIST AI RMF (Govern, Map, Measure, Manage), the EU AI Act risk tiers and prohibited practices, ISO/IEC 42001, OECD principles, GDPR automated decision-making, vendor AI risk and model inventories.
• Day 4: Full timed mock exam (60 questions, 60 minutes). Score it and list every miss by domain.
• Day 5: Drill your two weakest domains, then a second full mock. You want 80%+.
• Exam day −1: Light review only — the cheat sheet, the OWASP LLM list, and the framework names. Stop early. Sleep.

Cheat sheet: attack → the control that BEST stops it

A big share of SecAI+ questions ask for the best control. The winning answer is usually the one enforced outside the model, because anything the model is told can be talked around:

Watch for the classic trap: "add a stronger instruction to the system prompt." It's almost never the best answer, because a prompt rule is exactly what an injection overrides.

The 2-week plan (experienced track)

Same structure, compressed. Take each domain's quiz first and only study what you miss.

• Days 1–2: Basic AI concepts — skim what you know, focus on training terms (epochs, quantization, pruning), data provenance, and RAG/vector stores.
• Days 3–6: Securing AI systems: OWASP LLM Top 10, model attacks, ATLAS and the MIT taxonomy, then controls. This is 40% of the exam — give it four full days and 40+ questions a day.
• Days 7–8: AI-assisted security and offensive AI.
• Day 9: Governance, risk and compliance. Practitioners most often lose points here, because framework and regulation names rarely come up in daily work.
• Day 10: Full timed mock #1. List misses by domain.
• Days 11–12: Drill weak domains.
• Day 13: Full timed mock #2. Book the real exam once you clear 80%.
• Day 14: Light review, early night.

The final 48 hours (both tracks)

• Two days out: last full mock. If it's 80%+, you're done learning — trust it.
• One day out: a 20-minute skim of the cheat sheet, the ten OWASP LLM risks, and the framework names (NIST AI RMF, EU AI Act, ISO/IEC 42001, MITRE ATLAS). No new topics, no full exams.
• Exam day: about a minute per question is all you get. If a performance-based item or long scenario eats more than two minutes, flag it, move on, and come back with the time you've saved.

Before and after SecAI+

If you don't have a security foundation yet, start with Security+ — SecAI+ builds directly on it. If AI concepts are the gap, the foundational AWS AI Practitioner covers models, RAG and prompting from the ground up. After SecAI+, CySA+ deepens the SOC side. Not sure which direction fits? Take the two-minute which certification should I take quiz.

Frequently asked questions