Wirewalk Instituteof AI
sec-210 Intermediate Available

Prompt Injection and Agent Security

For security engineers, AI developers and red teamers who need to understand prompt injection deeply and build agents that stay safe when injection succeeds.

Level
Intermediate
Study time
12 hours
Lessons
17 in 5 modules
Exam
20 questions

About this course

Prompt injection sits at the top of the OWASP Top 10 for LLM Applications and drives the first entry of the OWASP Top 10 for Agentic Applications (ASI01 Agent Goal Hijack). It persists because a language model reads instructions and data through the same channel, so any content the model processes can try to steer it. This course explains the mechanism precisely, then follows payloads from documents, web pages, email, tool results, images and encoded text through to real impact: exfiltration through rendered links and tool calls, hijacked agent goals, and poisoned memory that persists across sessions. Public disclosures such as EchoLeak (CVE-2025-32711) are used as case studies.

You will evaluate detection approaches (input and output classifiers, spotlighting and delimiting, instruction hierarchy) and learn why adaptive attacks defeat them when used alone. The core of the course is architecture: the lethal trifecta of private data, untrusted content and an exfiltration path; plan-then-execute, dual-LLM and capability-based designs such as CaMeL; egress control; and confirmation steps people do not rubber-stamp. You finish by building an injection test corpus and automated probes with PyRIT, garak and promptfoo and tracking attack success rate as a regression metric. Broader agent identity, tool authorization and MCP hardening are covered in sec-220. The Wirewalk Institute certificate attests to completion and assessment of this course; it is not endorsed by OWASP, MITRE, NIST or any vendor named here.

What you will be able to do

  • Explain why prompt injection arises from mixing instructions and data in one context window
  • Distinguish direct injection, jailbreaks, indirect injection and multimodal or encoded payloads
  • Trace how an injected payload becomes data exfiltration, agent goal hijack or persistent memory poisoning
  • Evaluate classifiers, spotlighting and instruction hierarchy and explain their limits against adaptive attacks
  • Apply the lethal trifecta test and select architectural patterns that break at least one leg
  • Build an injection test corpus and automated probes and track attack success rate across releases

Syllabus

5 modules · 17 lessons · about 12 hours

Module 1: How Prompt Injection Works

The mechanism behind injection and the forms payloads take.
4 lessons
  1. Why Instructions and Data Share One Channel
  2. Direct Injection and Jailbreaks: Related but Different Problems
  3. Indirect Injection Through Documents, Web Pages, Email and Tool Results
  4. Multimodal and Encoded Payloads: Images, Audio, Unicode and Hidden Text

Module 2: From Injection to Impact

How payloads become exfiltration, hijacked goals and persistent compromise.
4 lessons
  1. Exfiltration Channels: Rendered Images, Links and Tool Calls
  2. Agent Goal Hijack (OWASP ASI01) and Tool Misuse
  3. Memory and Context Poisoning Across Sessions
  4. Case Studies from Public Disclosures, Including EchoLeak

Module 3: Detection and Its Limits

What filtering and prompt-level defenses buy you and why they are never enough alone.
3 lessons
  1. Input and Output Classifiers: What They Catch and Miss
  2. Spotlighting, Delimiting and Instruction Hierarchy
  3. Adaptive Attacks and Why Filters Alone Fail

Module 4: Architectural Defenses

Designs that limit damage when injection succeeds.
3 lessons
  1. The Lethal Trifecta: Private Data, Untrusted Content and an Exfiltration Path
  2. Plan-Then-Execute, Dual-LLM and Capability-Based Designs Such as CaMeL
  3. Egress Control, Output Rendering Rules and Human Confirmation

Module 5: Testing for Injection

Building repeatable tests and measuring progress.
3 lessons
  1. Building an Injection Test Corpus for Your Application
  2. Automated Probing with PyRIT, garak and promptfoo
  3. Measuring Attack Success Rate and Tracking Regressions

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 45 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by OWASP, MITRE ATLAS, Microsoft PyRIT, NVIDIA garak, or Promptfoo. Product and company names are trademarks of their respective owners.

More in AI Security: Securing AI Systems

All 13 courses in this track
sec-001 Free intro Available

Using AI Safely at Work: The Essentials

Follow one fictional US distributor through four everyday AI mistakes (a guarantor's credit file uploaded to a personal account, a customer letter promising a fee waiver that does not exist, a vendor email with hidden instructions for the assistant, and a note-taker that sent a customer the internal margin floor) and practice the decisions that prevent or contain each one in three short Decision Labs.

1 h · 7 lessons Free intro
sec-101 Basic Available

Using AI Safely: Data Handling for Every Employee

A plain-language course on handling data safely with AI tools: what happens to what you type or upload, how to classify before you paste, how connected apps and agents change the picture, and how to report a mistake quickly.

6 h · 15 lessons See price at checkout
sec-201 Intermediate Available

AI Security Fundamentals: The OWASP Top 10 for LLMs

A practitioner course on securing applications that use large language models, organized around the 2026 OWASP Top 10 for LLM Applications. Every risk comes with a hypothetical scenario, the root cause, and concrete defenses in code and configuration.

14 h · 18 lessons See price at checkout