AI Security Foundations
A free, 14-lesson introduction to assessing AI applications: prompt injection, retrieval, unsafe tools, agents, discovery, and evidence-based reporting.
Understand the system. Test the boundary. Verify the result.
AI Security Foundations follows the paths that connect instructions, retrieved content, tools, and agents to application behavior. Learn to distinguish a surprising answer from a demonstrated security consequence.
The public syllabus introduces the course. Written lessons, companion videos, exercises, references, and quizzes are available through the academy learning flow after free enrollment.
Practical assessment, grounded in evidence.
Recorded local demonstrations use synthetic data and real AI application workflows. Study the theory, follow the companion videos, and test your understanding. No advanced machine-learning background is required.
What the course covers.
Follow the trust boundaries that determine what an AI application can read, share, or change.
AI Application Security
Architecture, attack surfaces, common failures, prompt injection, and the context supplied to language models.
RAG and Data Leakage
Retrieval permissions, sensitive context, untrusted sources, and the difference between plausible output and verified exposure.
Tools and MCP
Tool permissions, unsafe invocation, connector metadata, and receiver-side verification of requested actions.
Agents and Discovery
Agent workflows, protocol boundaries, peer discovery, DNS metadata, and multi-agent failure chains.
Testing Workflow
Scope, hypotheses, repeatable evaluations, approval checks, and distinguishing failures from inconclusive results.
Evidence and Reporting
Traceable artifacts, reproducible findings, impact, remediation, retesting, and a report-based final capstone.
Your 14-lesson learning path.
Move from application fundamentals to connected agents, practical testing, and a final evidence-based assessment.
What Is AI Security?
Distinguish a model from its surrounding application.
Why AI Applications Are Different
Explain how natural-language data can resemble instructions.
AI Application Attack Surface
Map direct and indirect inputs.
Common AI Security Failures
Distinguish major failure mechanisms.
Prompt Injection Basics
Distinguish direct and indirect prompt injection.
How LLM Applications Build Context
Identify the sources assembled into a model request.
RAG and Data Leakage Basics
Trace retrieval into model context.
Unsafe Tool Use: When AI Assistants Can Take Actions
Trace model proposals into executable tool calls.
Agents and Why They Increase Risk
Explain the action-observation loop.
MCP Basics and Security Risks
Distinguish MCP host, client, and server responsibilities.
AI Agent Protocols and Discovery
Separate discovery, identity, and authorization.
Multi-Agent Systems and Failure Chains
Trace role handoffs and shared context.
AI Security Testing Workflow and Tools
Define measurable hypotheses before selecting tools.
Evidence, Reporting, Controls, and Final Capstone
Write evidence-supported conclusions with explicit limits.
Designed for focused progression.
Each lesson connects the security reasoning with a companion video and a five-question knowledge check.
Theory Lessons
Written explanations, objectives, concept diagrams, practical exercises, and primary-source references.
Companion Videos
Sketch explanations and selected recorded walkthroughs of local AI application workflows.
Applied Exercises
Practice reasoning about trust boundaries and preserve evidence. Finish with a self-assessed capstone report.
Knowledge Checks
Five questions per lesson, answer explanations, an 80% passing score, and retries.
What you will be able to do.
Map the attack surface
Identify where trust changes across user input, context, retrieval, tools, discovery, and agent handoffs.
Verify the consequence
Connect controlled inputs and tool requests to receiving-service records. Separate observed behavior from proven impact.
Report reproducible results
Document scope, conditions, evidence, limitations, remediation, and retest results without inventing a vulnerability.
Build your foundation in AI security.
For students, junior security practitioners, application security testers, and developers assessing AI-enabled applications. Basic familiarity with web applications is helpful.
Registration and enrollment are free. Demonstrations use local synthetic environments; paid operator-course lab access is not included. The final capstone is a self-assessment, not a certification exam.
Course Content
