A curated, always-current directory of free and self-hostable AI/LLM security labs, CTFs, competitions, bug bounties, tools, and references.
The AI Security Resource Hub is a single place to find hands-on material for learning and practicing AI/LLM security: prompt injection, jailbreaks, agent and MCP exploitation, RAG attacks, guardrail bypass, and more. Every entry is free to use or self-hostable. Paid courses and training subscriptions are intentionally left out.
Live, interactive site: https://arcanum-sec.github.io/ai-sec-resources/
| Category | Count | Description |
|---|---|---|
| Labs | 71 | Hands-on labs, CTFs, and vulnerable apps, including MCP and agent targets |
| Competitions | 7 | Gamified arenas and jailbreak/red-team competitions |
| Bug Bounties | 5 | Programs that pay for AI/LLM vulnerabilities |
| Tools & Utilities | 13 | Scanners, frameworks, and payload generators |
| Text Resources | 7 | Taxonomies, questionnaires, benchmarks, and references |
| Total | 103 |
Hosted challenges and self-hostable vulnerable apps covering prompt injection (direct and indirect), jailbreaking, data exfiltration, RAG poisoning, guardrail bypass, and agent/MCP exploitation. Includes staples like Gandalf, HackMerlin, GPT Prompt Attack, PortSwigger's Web LLM labs, and Prompt Airlines, alongside self-hosted suites such as the Black Hills AI-CTF, AISecOps Labs, LLMGoat, Bishop Fox's Local LLM CTF, and the Damn Vulnerable MCP/agent projects.
Ongoing and recurring competitions and arenas: HackAPrompt 2.0, Gray Swan AI Arena, RedTeam Arena, Pangea AI Escape Room, LLMail-Inject, the SaTML LLM CTF, and Alignment Arena.
Official AI vulnerability programs: Anthropic, OpenAI, Google Gemini, Mozilla 0din, and huntr (Protect AI).
Red-team and testing frameworks and scanners, including Praetorian's Augustus and Julius, NVIDIA garak, Microsoft PyRIT, Promptfoo, Giskard, CyberArk FuzzyAI, DeepTeam, NeMo Guardrails, and P4RS3LT0NGV3.
The Arcanum Prompt Injection Taxonomy, AI Pentest Questionnaire, and AI Security Ecosystem, plus community awesome-lists, JailbreakBench, and the seminal indirect-injection demos.
Contributions that add free or self-hostable AI security resources are welcome. There are two ways to suggest something:
Just open an issue. Point us at the resource (a link is enough, more context is better) and we will review it and add it ourselves, crediting you in the CHANGELOG. This is the easiest path if you do not want to touch the code.
Or open a pull request:
- Fork the repository.
- Add an object to the
RESOURCESarray inindex.html(cat,title,meta,url,action,desc,features, optionaltags). Counts and filters update automatically. - Verify the link is live and the resource is genuinely AI/LLM security material.
- Open a pull request.
Either way, please keep it to free or self-hostable resources. No paid courses or training subscriptions.
See CHANGELOG.md. Latest release: v2.0, a full redesign and content overhaul (79 resources).
For educational and authorized security testing only. Always get proper authorization before testing any AI system or application.
Curated by Jason Haddix, Arcanum Information Security. Website: https://arcanum-sec.com
If this hub is useful, please consider starring the repository.