🧭 Lesson Overview

If you have studied for or hold a security credential — Security+ or equivalent — you already know how to govern, assess risk, build a security program, and manage incidents. That knowledge is not wrong. It is incomplete.

The systems now entering enterprise environments do not behave like traditional software. They learn from data. They produce probabilistic outputs. They can be manipulated through their inputs in ways that have no parallel in SQL injection or buffer overflows. They drift in production even when nobody changes a line of code. They can take autonomous actions with real-world consequences. And the regulatory landscape around them is entirely new.

Today we establish the conceptual bridge: what Security+ covers, what it stops short of, and exactly where AI cybersecurity must extend it. By the end of this lesson you will be able to explain — to an executive, a regulator, or a skeptical peer — why traditional infosec principles alone are insufficient for AI-enabled enterprises, and what AAISM adds on top.

🎓 Anchoring Analogy: The Pilot's Type Rating

An airline pilot earns an Airline Transport Pilot License (ATPL) — the top-level certification that proves they can fly. But every new aircraft type requires a separate "type rating" because each aircraft has its own avionics, its own failure modes, its own emergency procedures. The pilot does not unlearn how to fly; they extend their qualifications to a new platform.

Security+ is the ATPL. AAISM is the type rating for AI. The foundational skills — governance, risk management, program development, incident response — transfer completely. What changes is the aircraft.

Day 1 Learning Objectives

By the end of this lesson you will be able to:

  1. Explain why traditional cybersecurity certifications alone are insufficient for AI-enabled enterprises and where AAISM extends them.
  2. List the six implicit assumptions in traditional security that AI breaks.
  3. Name the three AAISM domains and their percentage weights on the AAISM exam.
  4. Describe the course map — AI as target vs. AI as weapon — and the four major frameworks covered.
  5. Articulate at least three ways AI systems change the information security threat model.
A
Lecture & Structured Exploration — The Conceptual Bridge
⏱ ~2 hours

🏛️ Module 1 · Why Security+ Alone Isn't Enough

The Security+ body of knowledge covers a broad set of foundational cybersecurity domains: threats and vulnerabilities, cryptography, identity and access management, architecture, implementation, and incident response. These domains remain correct. They are also incomplete when the assets being governed include systems that learn from data, generate novel outputs, and can be manipulated through their inputs in ways that have no parallel in traditional software.

The Security+ Foundation — and Its Boundaries

🛡️ Security+ — Core Domains
  • ⚠️
    Threats & Vulnerabilities — identify, classify, and respond to threat actors and attack types
  • 🔐
    Cryptography & PKI — encryption, hashing, certificates, key management
  • 🪪
    Identity & Access Management — authentication, authorization, directory services
  • 🏗️
    Architecture & Implementation — network security, cloud, endpoint, and application hardening
  • 🚨
    Incident Response & Governance — IR lifecycle, risk management, compliance frameworks
Implicit assumption: deterministic systems, stable inputs/outputs, code-centric threats, known CVEs and signatures
🤖 AAISM — Three Domains
  • 🏛️
    D1 AI Governance & Program Management (31%) — AI-specific policy, data and model lifecycle, AI incident response
  • ⚠️
    D2 AI Risk Management (31%) — AI-specific threats, vulnerabilities, vendor and supply chain
  • 🔬
    D3 AI Technologies & Controls (38%) — security architecture, lifecycle controls, monitoring, ethics & trust
New assumptions: probabilistic behavior, adversarial inputs, training data as attack surface, continuous model change
⚠️ Eligibility Note
This course is grounded in Security+ foundational knowledge and is designed as the stepping-stone pathway toward advanced AI security governance. Professionals who continue to CISM or CISSP, then pursue AAISM, will find this course laid the conceptual groundwork. The core principle holds: you cannot competently govern AI security without first understanding how to govern security in general.

The Three ISACA AI Credentials — Knowing Which One Fits

AAISM is one of three new ISACA "Advanced in AI" credentials, each aimed at a different professional role. Understanding the distinctions helps security professionals make the right next-credential decision.

Credential Audience Prerequisite Focus
AAISM — Advanced in AI Security Management Security managers, CISOs, security architects CISM or CISSP
Pathway: Security+ → this course → CISM/CISSP → AAISM
Governing and securing enterprise AI
AAIA — Advanced in AI Audit Auditors, assurance professionals CISA or equivalent Auditing AI systems and controls
AAIR — Advanced in AI Risk Risk and compliance professionals CRISC or equivalent Enterprise AI risk management
💡 Bridge Concept

For a security professional, the learning pathway runs: Security+ establishes the foundational vocabulary → advanced certifications deepen governance expertise → AAISM adds the AI-specific layer on top. Think of Security+ as the operating system; this course extends it toward the AI security skills that the field now demands.

Module 2 · Six Things That Are Genuinely New in AI

Security professionals often assume they can extend existing mental models to cover AI. In many cases they can — but in six specific dimensions, AI systems behave in ways that require genuinely new thinking. Click each card to flip it and see the security implication.

🎲
Non-determinism
Click to flip
Security Implication
The same input can produce different outputs. Traditional testing assumes repeatability. AI systems break that assumption. Validation, monitoring, and incident triage all need new approaches. You cannot "test to completion."
📊
Data as Code
Click to flip
Security Implication
Training data becomes the system's behavior. Poisoned data is not just a confidentiality issue — it is a control-plane attack. Data lineage and provenance become security requirements, not nice-to-haves.
💬
Prompts as Attack Surface
Click to flip
Security Implication
Natural language is now a control plane. Instructions and data flow through the same channel — the model can't always distinguish them. Prompt injection is genuinely novel and has no clean traditional analogue.
🌊
Model Drift
Click to flip
Security Implication
Systems degrade silently over time. Production data shifts; model performance decays even when nothing in the code changed. Continuous monitoring of model behavior — not just availability — is required.
🤖
Agentic Action
Click to flip
Security Implication
Models can now take action, not just answer. When an AI agent has tool access — email, code execution, payments — the blast radius of a single compromised prompt expands dramatically. Excessive agency is a top OWASP LLM risk.
⚖️
Regulatory Novelty
Click to flip
Security Implication
New regulators, new categories, new fines. EU AI Act, state-level US laws, sector regulators, and emerging frameworks impose obligations that don't map cleanly to GDPR or HIPAA. Provider vs. deployer is a new distinction with legal teeth.
🔍 Why This Matters for Governance
Each of these six dimensions requires policy language, risk assessment methods, or control types that simply do not exist in a traditional infosec program. The goal of AI-103 is to build those policy constructs, risk frameworks, and controls — layer by layer — over the 15 days of this course.

🗺️ Module 3 · Course Map — AI as Target & AI as Weapon

AI-103 approaches AI cybersecurity from two angles that together cover the full scope of what a security professional must understand: AI as a target (how adversaries attack AI systems you build or deploy) and AI as a weapon (how adversaries weaponize AI to attack your organization). Both are real, both are growing, and the controls for each are partially — but not completely — overlapping.

🛡️
NIST AI RMF
Risk management framework. Govern, Map, Measure, Manage. Days 1–4.
🏛️
ISO/IEC 42001
Certifiable AI management system. AIMS analogue to ISO 27001. Days 2–5.
⚖️
EU AI Act
Binding regulation. Risk tiers, fines, extraterritorial reach. Day 5.
🔓
OWASP LLM Top 10
Practitioner threat reference. 2025 edition. Days 7–8.
🎯
AI as Target
Attacks on models and data. Poisoning, evasion, extraction. Days 7–9.
⚔️
AI as Weapon
Deepfakes, AI phishing, malware, influence operations. Day 10.
🔧
Secure Architecture
Defense in depth, AIBOM, model cards, lifecycle controls. Day 11.
🚨
AI Incident Response
NIST SP 800-61 adapted for AI. Red teams, tabletops. Days 14–15.

✏️ Activity A · Objective Mapping Exercise

Activity Map Course Objectives to Traditional Cybersecurity Principles ⏱ 25 min individual + 5 min share-out

For each course learning objective below, use the dropdown to select which traditional Security+ domain it maps to most closely, then decide whether it represents something Security+ already covers fully, partially covers, or doesn't cover at all. This exercise surfaces exactly where the AI extension is needed.

Course Learning Objective Closest Security+ Domain Security+ Coverage
Share-Out Prompt (5 min): Which objective was hardest to map? Which Security+ domain gets the most AI extensions? What does that tell you about where AI changes security practice most?
B
Deep Dive & Lab — AAISM Domains and the Security Gap
⏱ ~2 hours

🤖 Module 4 · AAISM Domains — The AI Security Extension

AAISM organizes AI security management into three domains. Understanding these domains serves two purposes: it maps to the AAISM exam content outline, and — more importantly for this course — it provides a conceptual architecture for everything we will build over the next 15 days.

Domain 1 — AI Governance & Program Management (31%)

This domain covers everything required to establish, operate, and maintain a formal AI security program within an enterprise. It extends beyond Security+'s governance and compliance fundamentals to add AI-specific requirements: AI steering committees, AI use case registers, and AI incident response annexes that Security+ does not address.

AAISM D1 Topic What It Covers Security+ Parallel / AI Gap
AI Steering Committee & Charter Cross-functional governance body with defined authority, cadence, escalation paths for AI-specific decisions Security+ covers basic governance concepts — AI steering committee is a new structure not found in Security+
AI Acceptable Use Policy Data classification rules for AI inputs, prohibited use cases, approval pathways, shadow AI handling Security+ covers policies and procedures at a conceptual level — AI AUP requires entirely new clauses Security+ never anticipated
AI Use Case Register Inventory of all approved AI deployments with owner, risk tier, review date — the "asset inventory" for AI Security+ covers asset management fundamentals — AI requires a new register format tied to behavioral risk, not just presence
Data & Model Lifecycle Governance Provenance, lineage, AIBOM, model cards — governance of the full lifecycle from training to retirement Security+ covers data classification at a foundational level — lifecycle governance of training data is new and requires AI-specific controls
AI Incident Response Annex Extension of NIST SP 800-61 IR lifecycle to cover AI-specific incidents: hallucinations, prompt injection, model poisoning Security+ covers the IR lifecycle — but AI incidents have novel detection, analysis, and remediation steps beyond standard playbooks

Domain 2 — AI Risk Management (31%)

This domain covers the identification, assessment, and treatment of AI-specific risks. Security+ introduces risk management concepts — but AI introduces risk categories that do not map cleanly onto CIA-triad thinking familiar from Security+.

AI Risk Category Definition Traditional Analogue?
Model Risk The model itself behaves incorrectly — hallucinates, drifts, fails on edge cases, performs worse on subpopulations Partial — software defects are analogous, but non-determinism makes model risk fundamentally different
Data Risk Training, fine-tuning, or query data creates problems — poisoned, biased, leaking, copyrighted, or regulated Partial — data security exists, but data-as-control-plane is new
Adversarial Risk Deliberate attacks targeting the AI system — prompt injection, evasion, extraction, inversion, poisoning Partial — adversarial attacks exist in traditional security, but the mechanisms are entirely different
Third-Party / Supply Chain Risk Dependencies on models, data, or services you don't control; inherited vulnerabilities from the AI supply chain Partial — supply chain risk exists, but the AI supply chain has unique opacity and layering
Ethical & Societal Risk System causes harm even when functioning as designed — discrimination, displacement, deception Minimal — traditional infosec rarely addresses designed-in harm; AI governance requires it
Regulatory & Legal Risk Deployment violates binding rule — EU AI Act, sector regulation, IP law, consumer protection Partial — regulatory risk is known; AI-specific regulation is new and rapidly evolving

Domain 3 — AI Technologies & Controls (38%)

The largest AAISM domain covers the technical architecture and operational controls required to secure AI systems. This is where threat landscape knowledge (OWASP, ATLAS), architecture design (defense in depth for AI), and operational monitoring intersect. Day 11 builds the architecture; Days 7–9 build the threat model. For today, we establish the vocabulary.

🔬
Security Architecture
Defense in depth across network, data, application, and model layers for AI systems.
📋
Lifecycle Controls
Controls from data collection through training, deployment, monitoring, and retirement.
📡
Behavioral Monitoring
Monitoring model behavior — not just availability — for drift, anomalies, and adversarial signals.
⚖️
Ethics & Trust
Fairness, explainability, and accountability controls aligned to NIST AI RMF trustworthiness characteristics.
🔓
Threat Controls
Input/output filtering, system prompt hardening, least privilege for agents, content filters.
🔍
Red Team & Testing
Adversarial testing — prompt injection, jailbreaks, poisoning, extraction — appropriate to system risk.
🔑 Key Insight: Why 38%?

Domain 3 carries the largest exam weight because technical controls are where AI security fails most visibly — and because the skills do not transfer cleanly from traditional security. A Security+-certified professional who can design a firewall ruleset does not automatically know how to design a prompt injection defense, an AI model isolation architecture, or an MLOps monitoring pipeline. These are genuinely new skills.

🔬 Activity B · Think-Pair-Share: Where Does the Gap Live?

Think-Pair-Share AI Security Gap Analysis — Community College Context ⏱ 30 min

Moraine Valley Community College is deploying three AI systems this semester:

  1. AI tutoring chatbot for developmental math students — trained on MVCC course materials, deployed on the student portal.
  2. AI-assisted grading tool — helps faculty evaluate written assignments, surfaces suggested scores and feedback.
  3. Predictive analytics dashboard — flags at-risk students for advising intervention based on attendance, LMS engagement, and grade data.
Instructions:
  1. Think (5 min): For each system, identify one traditional cybersecurity control that applies and one challenge that is genuinely new (not covered by traditional infosec). Write your answers individually.
  2. Pair (10 min): Compare with a partner. Where did you agree? Where did you reach different conclusions?
  3. Share (10 min): Class share-out. Instructor captures on whiteboard: traditional controls column vs. new AI challenges column.
  4. Debrief (5 min): Which of the six "genuinely new" assumptions from Module 2 appear most in your analysis?

Instructor note: The predictive analytics dashboard is designed to surface the ethical/societal risk dimension — at-risk student flagging has real equity implications that traditional CIA triad thinking doesn't capture.

📝 Assessment Artifact

📋 Day 1 Assessment Artifact

One-Page Reflection: "Three Ways AI Breaks Traditional InfoSec Assumptions"

Write a focused one-page (400–500 word) reflection addressed to a hypothetical CISO colleague who holds a Security+ and is skeptical that AI requires new governance structures: "We've secured systems before. Why is AI fundamentally different?"

Your reflection must address:

  • Three specific ways AI systems break implicit assumptions in traditional information security — choose from the six covered in Module 2, but go deeper than the card text. Use a concrete example for each.
  • One AAISM domain that you believe addresses the most critical gap — and why you chose it over the other two.
  • One analogy of your own (not the pilot/type rating analogy from the lesson) that explains the Security+→AAISM credential pathway to a non-technical stakeholder.

Format: Plain prose, no bullet lists (write it as a professional memo, not lecture notes). You will use this reflection as a reference point on Day 15 when you build your capstone AI Security Program.

Grading: Daily assignment category (25% of course grade). Evaluated on specificity of AI security gap identification, accuracy of AAISM domain characterization, and quality of original analogy.

Day 1 Knowledge Check

Test your mastery of Day 1 content. Each question provides immediate feedback with an explanation. This is formative — it does not affect your grade, but it predicts how well you will do on the Week 1 quiz.

Day 1 Knowledge Check — 5 Questions Score: 0 / 5
Q1. A Security+-certified security manager says: "We already handle security risk — just add an 'AI' section to the existing risk register." Which response BEST explains why this is insufficient?
Q2. Which of the following BEST describes why "training data as attack surface" is a genuinely new security concept?
Q3. AAISM's Domain 3 (AI Technologies & Controls) carries the highest exam weight at 38%. Which explanation BEST justifies this weighting?
Q4. A security architect discovers that a recently deployed customer-service chatbot has begun producing different answers to the same question depending on how the question is phrased — even when the correct answer should be identical. Which "genuinely new AI assumption" does this BEST illustrate?
Q5. Which statement MOST accurately describes the relationship between Security+ and AAISM?