PromptisePromptise
Why PromptiseUse casesRoadmapDocs
GitHubGet started
Promptise - AI Framework LogoPromptise

The foundation layer for agentic intelligence. Build, secure and operate autonomous AI systems with Promptise Foundry.

pip install promptise

[01] Foundry

  • MCPcast
  • The Promptise Agent
  • Reasoning Engine
  • MCP
  • Agent Runtime
  • Prompt Engineering
  • Execution Engine
  • Agent Identity

[02] Resources

  • Documentation
  • GitHub
  • Guides
  • Learning Paths
  • Questions

[03] Company

  • About
  • Terms of Service
  • Privacy Policy
  • Cookie Policy
  • Subprocessors

© 2026 Promptise by Manser Ventures. All rights reserved.

Open source · Python

← Learning paths> LLM Security· featured
Advanced path

LLM Security Course: Prompt Injection to Red Teaming

Start the pathAll paths

A free LLM security course for developers and security teams, from how LLM attacks work to defending production systems. You learn the main attack vectors (prompt injection, data leakage, unsafe output), how to structure prompts and validate input and output, how to red-team your own system, and how to monitor and respond when something goes wrong. Each lesson is a hands-on guide with examples you can run.

16
Lessons
5.8 h
Reading time
Advanced
Level
Free
Cost

The lessons

16 lessons

Read the lessons in order; they are arranged so each one prepares the next. Every lesson is a guide, so you can also open any of them on its own.

  1. [01]LLM Attack Surface: Vectors, Risks and the OWASP Top 10The LLM attack surface explained: user apps, integrations, APIs, data stores, insiders and supply chain, the common attack vectors, and the OWASP Top 10.Intermediate · 11 minOpen →
  2. [02]What Is Prompt Injection? How It Works, With ExamplesPrompt injection explained: how direct and indirect attacks hijack a language model, a real example of it working, and the defenses that reduce the risk.Beginner · 9 minOpen →
  3. [03]Secrets in Prompts: System Prompt Leakage and Safe PatternsWhy API keys and access rules in a system prompt leak, a real extraction attempt, and patterns that keep secrets in tools, code and secret managers.Intermediate · 9 minOpen →
  4. [04]Stop Secrets Leaking Into LLM Logs and TracesHow API keys and tokens end up in LLM logs and traces, and how to redact them from tool results, observability and log lines before anything is stored.Intermediate · 18 minOpen →
  5. [05]Leaked LLM API Key? Rotate and Audit in 30 MinutesLeaked an OpenAI, Anthropic or other LLM API key? A 30-minute checklist: revoke and rotate, check usage and spend, find the source, and stop the next leak.Beginner · 8 minOpen →
  6. [06]Secure Prompt Engineering: Separate Instructions From DataSecure prompt engineering: keep untrusted text as data with message roles, delimiters, spotlighting, rules and output schemas. A real before and after.Intermediate · 9 minOpen →
  7. [07]LLM Input Validation: Why Regex Fails and What WorksValidate LLM input in layers: limits, Unicode normalization, field rules and a scope classifier. Real examples of regex failing and what catches attacks.Intermediate · 9 minOpen →
  8. [08]LLM Output Validation: Check Responses Before You Use ThemValidate LLM output before you act on it: schema, grounding and consistency checks, safe handling of links, SQL and HTML, and when to retry or escalate.Intermediate · 10 minOpen →
  9. [09]RAG Security: Stop Poisoned Documents and Prompt InjectionSecure a RAG pipeline: stop poisoned documents and indirect prompt injection, filter retrieval by permission, mark context, check answers. Real examples.Advanced · 10 minOpen →
  10. [10]Hidden Text Prompt Injection: Sanitize Pages for AI AgentsStop hidden text prompt injection: strip CSS-hidden text, comments and invisible characters from web pages, scan what is left, and limit agent tools.Intermediate · 18 minOpen →
  11. [11]LLM Red Teaming: How to Test Your Own LLM ApplicationRed-team your own LLM app: define what must never happen, write probes per attack type, run a harness with canaries, fix findings and keep them in CI.Advanced · 11 minOpen →
  12. [12]LLM Hardening: Constrain Model Output When Prompts FailHarden LLM behavior beyond the prompt: structured outputs, constrained decoding, output checks, safety classifiers and sampling limits, with real examples.Advanced · 11 minOpen →
  13. [13]LLM Security Monitoring: Detect Attacks With Audit LogsMonitor LLM apps for attacks: what security events to log, which signals mean trouble, detection rules in Python, and alerting without alert fatigue.Intermediate · 10 minOpen →
  14. [14]LLM Supply Chain Security: Models, Packages, Data and ToolsLLM supply chain security as a checklist: inventory models, data and tools, pin and verify hashes, load safe formats, check provenance and plan fallbacks.Intermediate · 9 minOpen →
  15. [15]uv Dependency Cooldowns and Pins for Python AI AppsSet a uv dependency cooldown with exclude-newer, pin exact versions in a hashed lockfile and install only from it, so a poisoned release never reaches you.Intermediate · 8 minOpen →
  16. [16]LLM Security Checklist for Production SystemsAn LLM security checklist for production: prompts and secrets, injection, RAG access, output handling, tools, dependencies, limits, monitoring, sign-off.Intermediate · 10 minOpen →
Start lesson 01

Before you start

  • To get the most from this path, you should know the basics of how LLMs work. That means the input-processing-output cycle and how prompts act as instructions. Familiarity with prompting concepts like system instructions, user input, and role-based prompts is important, and ideally you've already spent time writing and testing prompts in a real LLM tool like ChatGPT or Claude. On the technical side, comfort reading and writing code (or at minimum, following pseudocode) will help, along with some experience working with backend systems, APIs, and input validation patterns. You should also be comfortable with the basics of logging, monitoring, and alerting concepts.

Skills you build

  • Identify and articulate the three core LLM attack vectors (input, output, context)
  • Recognize prompt injection vulnerabilities and design injection-resistant prompts
  • Audit systems for leaked secrets and implement safe secret injection patterns
  • Design multi-layer prompt architectures that resist manipulation
  • Build input validation pipelines specific to LLM risks
  • Verify LLM outputs and detect hallucinations without over-engineering
  • Secure retrieval-augmented generation (RAG) systems against poisoning and injection
  • Red-team your own systems to find vulnerabilities before attackers do
  • Harden model behavior at inference time using multiple techniques
  • Design monitoring and alerting systems for LLM security incidentsEvaluate LLM providers and models for security posture
  • Apply a productioUnderstand EU AI Act requirements and compliance implicationsn-ready security checklist before launch
  • Make informed trade-offs between security, capability, and performance

By the end, you can

  1. 01Understand the LLM security landscape and where real risks hide
  2. 02Be able to design and audit prompts for safety
  3. 03Know how to defend against the most common LLM attacks
  4. 04Understand when prompting is enough and when you need additional hardening
  5. 05Be able to build systems that detect and respond to attacks
  6. 06Know how to evaluate and manage LLM providers safely
  7. 07Have a checklist and process for shipping secure LLM systems
  8. 08Understand regulatory requirements (EU AI Act) and how to approach compliance
  9. 09Be confident explaining LLM security trade-offs to stakeholders
  10. 10Have practical, implementable techniques you can apply immediately

Ready when you are.

16 lessons, in order. Each opens with the full list beside it, so you always know where you are and what comes next.

Start the pathThe guide library