Skip to main content

AI Security

Reference material on artificial intelligence and machine learning security: offensive techniques against AI systems and defensive controls to protect them.

CAUTION

Authorized use only. All material is provided for educational purposes and authorized testing.
See ../SECURITY.md for the full policy.

Scope​

This category covers the attack surface introduced by AI and ML systems: large language models (LLMs), retrieval-augmented generation (RAG) pipelines, autonomous agents, the model supply chain, and the infrastructure that serves them. Each topic is written in a two-track format: offensive (how the technique works, how to test for it) and defensive (how to prevent and detect it).

Format​

Every file follows the two-track structure:

  • Offensive track - reconnaissance, attack vectors, exploitation recipes, post-exploitation.
  • Defensive track - preventive controls, detection controls, verification.

Section headers have no suffix - the section names themselves make the track clear. See WRITEUPS/ for HTB walkthroughs and APPLICATIONS/ for web application security.

Subcategories​

#SubcategoryFilesFocus
01OWASP-LLM-Top-10-202610OWASP LLM Top 10 (2026) threats and mitigations
02RAG-and-Embeddings4RAG pipelines, vector databases, embedding attacks
03Agents-and-MCP4Autonomous agents, Model Context Protocol (MCP) security
04AI-Supply-Chain4Model provenance, dataset poisoning, dependency attacks
05OSAI-Prep3Offensive Security AI (OSAI) certification preparation
  • APPLICATIONS/ - web application security, including API security.
  • INFRASTRUCTURE/ - infrastructure attack and defense (AD, Windows, Linux, network).
  • CLOUD/ - cloud security (planned).

Authorized Use Only​

The techniques in this category are for authorized security testing, CTF environments, and educational purposes only. Do not use them against systems you do not own or have explicit written permission to test. See SECURITY.md in the repository root.

References​