AI security learning spans several responsibilities. A leader may need a map of the risks, a developer may need to understand trust boundaries, and a security tester may need an organized way to evaluate model behavior. These courses address those different starting points.
The selection emphasizes understanding and defensive assessment of AI applications. Enterprise-agent design receives a separate option because access, memory and tool use add their own architectural questions.
Table of Contents
Compare the free courses
| Course or lesson | Best fit | Main focus |
|---|---|---|
| Introduction to Security in the World of AI | Understanding the main security questions | AI risks across data, models and applications |
| AI Security: Prompt Injection and Zero Trust | Recognizing application trust boundaries | Prompt injection, privileges, data and delegation |
| AI Red Teaming: Prompt Injection and PyRIT | Studying organized defensive testing | Model behavior, test patterns and automation |
| Secure Enterprise AI Agents | Designing agent access and information boundaries | Identity, state, memory and data perimeters |
Recommended courses and what makes each useful
-
1. Introduction to Security in the World of AI
The Google introduction connects AI features with the data, models and actions around them. Its conceptual material is aimed at security and data-protection leaders.
Choose it when you need a first framework for discussing risks before a more technical implementation or assessment course.
-
2. AI Security: Prompt Injection and Zero Trust
IBM’s explainers examine risks around language-model applications and agents, including excessive privileges, exposed information and delegation. Trust boundaries and safeguards provide the central perspective.
Use it when the application’s connections and permissions need more careful reasoning. It is broader than a demonstration of one particular defensive product.
-
3. AI Red Teaming: Prompt Injection and PyRIT
The ten-part Microsoft series introduces direct and indirect prompt injection, single-turn and multi-turn testing, defenses and PyRIT automation.
Choose it when your learning goal is evaluating AI-system behavior through a structured testing workflow. Its assessment focus differs from the leader-oriented conceptual introduction.
-
4. Secure Enterprise AI Agents
The short Google course connects identity and access management with agent state, memory, data perimeters and conversation protection. Guardrails appear within the wider enterprise design.
Choose it when the application can use tools or interact with organizational information. The focused agent context makes it more specific than a general AI-risk overview.
Choose the security responsibility
For a first risk map, choose Security in the World of AI. For application boundaries, choose Prompt Injection and Zero Trust. For testing, choose AI Red Teaming. For enterprise-agent controls, choose Secure Enterprise AI Agents.
Frequently asked questions
Which course is most relevant to AI red teaming?
AI Red Teaming: Prompt Injection and PyRIT specifically develops testing patterns, model-behavior assessment, defenses and automation.
Which selection focuses on an agent’s access and memory?
Secure Enterprise AI Agents addresses identity, access, state, memory, data perimeters and conversation protection.
Questions & discussion
Share a useful question or correction. Comments appear after moderation. Please avoid personal or sensitive information.