shield

AI Security — Deep Dive

From threat landscape to production hardening. Each chapter: visual journey overview + under-the-hood deep dive.
Co-Created by Kiran Shirol and Claude
Core Topics OWASP Top 10 Guardrails Red Team LLM Security
home Learning Portal play_arrow Start Learning summarize Key Insights dictionary Glossary 14 chapters · Each with High Level + Under the Hood
Offense

Attack Surface & Threat Landscape

The attacks every AI practitioner must understand — from prompt injection to adversarial ML.
1
public
The AI Security Landscape
OWASP Top 10 for LLMs, MITRE ATLAS, the AI Incident Database, and the CIA triad applied to AI.
visibility High Level code Deep Dive
2
dangerous
Prompt Injection — The #1 Threat
Direct and indirect injection, the confused deputy problem, and real-world incidents.
visibility High Level code Deep Dive
3
lock_open
Jailbreaking & Guardrail Bypass
Crescendo attacks, many-shot jailbreaking, DAN role-play, and encoded payloads.
visibility High Level code Deep Dive
4
science
Data Poisoning & Training-Time Attacks
Sleeper agents, PickleRAT supply chain attacks, safetensors, and model signing.
visibility High Level code Deep Dive
5
bug_report
Adversarial Machine Learning
FGSM, PGD, C&W attacks, evasion of safety classifiers, and transferability.
visibility High Level code Deep Dive
Defense

Guardrails, RAG, Agents & MCP Security

Securing the components of modern AI systems — from input filtering to tool-calling sandboxes.
6
filter_alt
Input Guardrails & Output Filtering
NeMo Guardrails, LLM Guard, canary tokens, PII detection, and layered defense.
visibility High Level code Deep Dive
7
search
Securing RAG Pipelines
CPA-RAG attacks, embedded threats, jamming attacks, RAGPart/RAGMask defenses.
visibility High Level code Deep Dive
8
smart_toy
Securing Agents & Tool Calling
AgentXploit, STAC tool chaining, WASM sandboxing, and capability-based access control.
visibility High Level code Deep Dive
9
hub
Securing MCP & External Integrations
MCPoison, tool poisoning, rug pull attacks, RSA manifest signing, and runtime guardrails.
visibility High Level code Deep Dive
Governance

Privacy, Red Teaming & Compliance

Data leakage, adversarial testing methodologies, and the regulatory landscape.
10
privacy_tip
Privacy, Data Leakage & Model Extraction
Membership inference, differential privacy, PII leakage, model stealing, GDPR/CCPA.
visibility High Level code Deep Dive
11
target
Red Teaming AI Systems
Garak, PyRIT, PromptFoo, MITRE ATLAS methodology, prompt fuzzing, and bug bounties.
visibility High Level code Deep Dive
12
gavel
AI Governance, Compliance & Risk
EU AI Act, NIST AI RMF, ISO 42001, model cards, incident response, and responsible disclosure.
visibility High Level code Deep Dive
Hardening

Secure Architecture & Production Hardening

Putting it all together — zero-trust patterns, defense in depth, and the full security stack.
13
architecture
Secure AI Architecture Patterns
Zero-trust for AI, layer separation, API gateways, rate limiting, and secrets management.
visibility High Level code Deep Dive
14
security
Production Hardening & Defense in Depth
The full security stack, continuous red teaming, incident response, and future outlook.
visibility High Level code Deep Dive
explore

Explore Related Courses

psychology
Agentic AI
Planning, Memory & Tool Use
search
RAG
Retrieval-Augmented Generation
hub
Model Context Protocol
Tool Integration & Standards
model_training
Fine-Tuning
Adapting Models to Your Data