Awesome AI Security - Project Overview
Purpose
This is a curated collection of AI/ML security materials and resources for pentesters, red teamers, and security researchers. The goal is to keep the list AI-focused, high-signal, well-categorized, and non-duplicated.
Project Structure
awesome-ai-security/
├── README.md # Main resource list (curated)
├── LICENSE # License
├── .claude/
│ └── skills/ # Claude skills (this directory)
└── ref/ # Reference notes (not curated)
├── my_collect.md # Personal collection
├── Awesome-AI-Security-1/
├── awesome-ai-security-2/
├── 模型安全/ # Model security notes
├── 渗透测试相关/ # Pentesting notes
└── 网络安全相关/ # Network security notes
README.md Format Convention
Heading Structure
- Top-level categories use
##.
- Subcategories use
### (e.g., inside AI Security & Attacks).
- Starter Pack uses bold bullets for sub-sections (e.g.,
- **CTFs / Practice**).
Link Format
- Use full URLs, one per bullet line.
- Add a short description in square brackets:
- https://... [Short description]
- Keep descriptions concise.
- Do not add the same URL in multiple places.
Example Entry
### Prompt Injection
- https://github.com/example/tool [Prompt injection detector]
Categorization Rules (How to Place a New Link)
- AI Security Starter Pack: CTFs, courses, blogs, newsletters, beginner resources.
- AI/LLM Guide: LLM fundamentals, tutorials, awesome lists.
- AI Security & Attacks: Prompt injection, adversarial attacks, poisoning, privacy, model security.
- AI Pentesting & Red Teaming: AI-powered pentesting tools, red teaming, MCP security tools.
- AI Security Tools & Frameworks: AI vulnerability detection, CVE analysis, OSINT, security libraries, TLS / fingerprint / bot signals (JA3 clients, site bot detection, automation hardening research—use only ethically and on authorized targets).
- AI Agents & Frameworks: Agent frameworks, formal methods / Lean agents (e.g. AI-assisted theorem proving orchestration), AI memory & long context (latent memory, recursive context, long-memory RAG), RAG stacks/collections, browser automation, MCP servers, agent sandboxes & isolation (policy-enforced runtimes, container/VM boundaries).
- AI Development & Training: Training frameworks, local models, uncensored models, prompts.
- AI Applications: Chat assistants, deep research, search engines, code analysis, web scraping, vision / domain apps (e.g. agricultural or specialized image understanding with LLMs).
- AI Image & Video: Image generation, video generation, TTS, face recognition.
- Benchmarks & Standards: AI safety benchmarks, threat frameworks, standards.
AI-Relevance Filter
Only include AI/ML-related resources. Do not add:
- Traditional security tools (unless AI-powered)
- Web3/blockchain tools (unless AI-related)
- General pentesting tools without AI integration
- Browser vulnerabilities, phishing tools, CVE collections (unless AI-analyzed)
Duplicate Policy
No duplicate URLs in README.md. If a link fits multiple categories, pick the primary one.
Contribution Checklist
- Check for duplicates in
README.md before adding.
- Verify the resource is AI/ML-related.
- Verify the link points to the canonical source (avoid low-value forks).
- Keep the description concise and useful.
- Put it into the most appropriate category.
- Prefer minimal changes over reformatting large sections.
Utilities Section
End of README.md includes Utilities (mixed): agent-facing CLIs, productivity, and mail/identity (e.g. self-hosted domain mail, encrypted P2P email) when they support ops or privacy around AI workflows—keep entries concise.
Data Source
For detailed and up-to-date resources, fetch the complete list from:
https://raw.githubusercontent.com/gmh5225/awesome-ai-security/refs/heads/main/README.md
Use this URL to get the latest curated links when you need specific tools, papers, or resources.
1---2name: awesome-ai-security-overview3description: Guide for understanding and contributing to the awesome-ai-security curated resource list. Use this skill when adding resources, organizing categories, or maintaining README.md consistency (no duplicates).4---56# Awesome AI Security - Project Overview78## Purpose910This is a curated collection of AI/ML security materials and resources for pentesters, red teamers, and security researchers. The goal is to keep the list **AI-focused**, **high-signal**, **well-categorized**, and **non-duplicated**.1112## Project Structure1314```15awesome-ai-security/16├── README.md # Main resource list (curated)17├── LICENSE # License18├── .claude/19│ └── skills/ # Claude skills (this directory)20└── ref/ # Reference notes (not curated)21 ├── my_collect.md # Personal collection22 ├── Awesome-AI-Security-1/23 ├── awesome-ai-security-2/24 ├── 模型安全/ # Model security notes25 ├── 渗透测试相关/ # Pentesting notes26 └── 网络安全相关/ # Network security notes27```2829## README.md Format Convention3031### Heading Structure3233- Top-level categories use `##`.34- Subcategories use `###` (e.g., inside `AI Security & Attacks`).35- Starter Pack uses bold bullets for sub-sections (e.g., `- **CTFs / Practice**`).3637### Link Format3839- Use full URLs, one per bullet line.40- Add a short description in square brackets: `- https://... [Short description]`41- Keep descriptions concise.42- Do not add the same URL in multiple places.4344### Example Entry4546```markdown47### Prompt Injection48- https://github.com/example/tool [Prompt injection detector]49```5051## Categorization Rules (How to Place a New Link)5253- **AI Security Starter Pack**: CTFs, courses, blogs, newsletters, beginner resources.54- **AI/LLM Guide**: LLM fundamentals, tutorials, awesome lists.55- **AI Security & Attacks**: Prompt injection, adversarial attacks, poisoning, privacy, model security.56- **AI Pentesting & Red Teaming**: AI-powered pentesting tools, red teaming, MCP security tools.57- **AI Security Tools & Frameworks**: AI vulnerability detection, CVE analysis, OSINT, security libraries, **TLS / fingerprint / bot signals** (JA3 clients, site bot detection, automation hardening research—use only ethically and on authorized targets).58- **AI Agents & Frameworks**: Agent frameworks, **formal methods / Lean agents** (e.g. AI-assisted theorem proving orchestration), **AI memory & long context** (latent memory, recursive context, long-memory RAG), RAG stacks/collections, browser automation, MCP servers, **agent sandboxes & isolation** (policy-enforced runtimes, container/VM boundaries).59- **AI Development & Training**: Training frameworks, local models, uncensored models, prompts.60- **AI Applications**: Chat assistants, deep research, search engines, code analysis, web scraping, **vision / domain apps** (e.g. agricultural or specialized image understanding with LLMs).61- **AI Image & Video**: Image generation, video generation, TTS, face recognition.62- **Benchmarks & Standards**: AI safety benchmarks, threat frameworks, standards.6364## AI-Relevance Filter6566**Only include AI/ML-related resources.** Do not add:6768- Traditional security tools (unless AI-powered)69- Web3/blockchain tools (unless AI-related)70- General pentesting tools without AI integration71- Browser vulnerabilities, phishing tools, CVE collections (unless AI-analyzed)7273## Duplicate Policy7475**No duplicate URLs in README.md.** If a link fits multiple categories, pick the primary one.7677## Contribution Checklist78791. Check for duplicates in `README.md` before adding.802. Verify the resource is AI/ML-related.813. Verify the link points to the canonical source (avoid low-value forks).824. Keep the description concise and useful.835. Put it into the most appropriate category.846. Prefer minimal changes over reformatting large sections.8586## Utilities Section8788End of `README.md` includes **Utilities** (mixed): agent-facing CLIs, productivity, and **mail/identity** (e.g. self-hosted domain mail, encrypted P2P email) when they support ops or privacy around AI workflows—keep entries concise.8990## Data Source9192For detailed and up-to-date resources, fetch the complete list from:9394```95https://raw.githubusercontent.com/gmh5225/awesome-ai-security/refs/heads/main/README.md96```9798Use this URL to get the latest curated links when you need specific tools, papers, or resources.
gmh5225