🎯 Working on an AI/LLM Security Engagement?

Use this handbook as your practical field guide throughout the entire assessment lifecycle—from initial reconnaissance against chatbots and AI-powered applications to advanced testing of prompt injection, jailbreak techniques, tool abuse, agent manipulation, and data exfiltration scenarios involving MCP, RAG, and computer-use systems.

Each chapter combines battle-tested attack payloads with clear guidance on expected success indicators, severity considerations, detection opportunities, and remediation recommendations. This enables both offensive and defensive teams to understand not only how an attack works, but also how to identify, mitigate, and prevent it.

The handbook covers the complete OWASP Top 10 for Large Language Model Applications, while also exploring emerging attack surfaces that extend beyond current industry frameworks.

Content is curated from a combination of public research, vendor disclosures, CVEs, academic papers, real-world incidents, and active red-team engagements, providing a comprehensive reference for modern AI security testing.

⚠️ For authorized testing only.

👉 Start Here — Open the Payload Library Master Index → PAYLOADS.md

Every payload on one page, grouped by attack class — copy-paste ready, full sets one click away. No digging through folders; it's all reachable from the master index.

For technique-level coverage with detection and mitigation, browse the chapters below.

Chapters

Foundations

OWASP LLM Top 10

Beyond OWASP

Frontier Vectors

Payload Library — by goal

Full one-page index with inline top picks: PAYLOADS.md

Repository Layout

AI-Pentest-Playbook/
├── PAYLOADS.md     ⭐ one-page payload index — start here
├── payloads/       the payloads, grouped by attack class
├── docs/           technique chapters (detection + mitigation)
├── scripts/        runner · burp-export · master-csv
├── README.md
└── CONTRIBUTING.md

💜 Found This Useful? Give Back.

This handbook gets sharper every time a real-world finding is distilled back into reusable knowledge.

🛠 Got a new payload, defence pattern, or attack surface that should be a chapter?

Most payload PRs merge within a day — new chapters and defensive notes are very welcome. A ⭐ helps other AppSec / red-team folks find the handbook.

🐛 Spotted something wrong, missing, or outdated? → Open an issue. Accuracy beats completeness.

Citation

@misc{aihackershandbook,
  author       = {4vanish and contributors},
  title        = {AI Hacker's Handbook: A playbook for pentesting AI chatbots and LLM-powered applications},
  publisher    = {GitHub},
  howpublished = {\url{https://github.com/4vanish/AI-Pentest-Playbook}}
}

Author

Built and maintained by Avanish Pathak.