📋 32 articles
💪 Intermediate
✨ Free · Hands-on
Defeat alignment training and safety filters. Many-shot jailbreaks, role-play exploits, encoding attacks, gradient-based suffixes, multilingual bypasses and the techniques researchers use to break Claude, GPT-4, Gemini and Llama.
✓ What you'll learn
- Many-shot jailbreaking
- Role-play & DAN attacks
- Encoding & ciphers (base64, ROT13)
- Gradient-based adversarial suffixes
- Multilingual safety bypasses
- Cross-model transfer attacks
Explore Track →
📋 53 articles
💪 Beginner to Advanced
✨ Free · Hands-on
The number-one LLM vulnerability in OWASP LLM Top 10. Direct injection, indirect injection through retrieved data, agentic workflow hijacking, API key theft via prompts, and bug-bounty-grade payloads that work on real production AI apps.
✓ What you'll learn
- Direct prompt injection
- Indirect injection (RAG, web fetch)
- Agentic workflow hijacking
- API key & credential extraction
- Filter & guardrail bypasses
- Bug bounty payloads & disclosure
Explore Track →
📋 168 articles
💪 Intermediate to Advanced
✨ Free · Hands-on
Full-spectrum LLM exploitation: model inversion, training-data extraction, embedding inversion, model theft via API queries, plugin architecture attacks, AI worms and the offensive security techniques that defined the 2025-2026 threat landscape.
✓ What you'll learn
- Model inversion & training-data extraction
- Membership inference attacks
- Insecure plugin architecture
- AI worms & self-propagating malware
- Code-assistant backdoor injection
- Deepfake-driven offensive security
Explore Track →
📋 61 articles
💪 All levels
✨ Free · Hands-on
The complete OWASP Top 10 for Large Language Models, mapped to real disclosed breaches, CVSS scoring, bug bounty payouts and a red-team assessment framework. One article per LLM-XX risk with hands-on exploit examples.
✓ What you'll learn
- LLM01 -- Prompt Injection (direct + indirect)
- LLM02 -- Sensitive Information Disclosure
- LLM03 -- Supply Chain (datasets, model registries)
- LLM04 -- Data and Model Poisoning
- LLM05 -- Improper Output Handling
- LLM06 -- Excessive Agency (LLM06 sub-types)
- LLM07 -- System Prompt Leakage (NEW in 2025)
- LLM08 -- Vector and Embedding Weaknesses (NEW in 2025)
- LLM09 -- Misinformation
- LLM10 -- Unbounded Consumption
Explore Track →
📋 90 days
💪 Beginner to Advanced
✨ Free · Hands-on
A 90-day structured curriculum from "what is an LLM" through to advanced red-team exercises. Day-by-day lessons, hands-on labs, daily missions and XP rewards. The fastest path to becoming a working AI security tester.
✓ What you'll learn
- Phase 1 (Days 1-15): LLM fundamentals & threat model
- Phase 2 (Days 16-30): Prompt injection in depth
- Phase 3 (Days 31-50): Jailbreaking & OWASP LLM 2025
- Phase 4 (Days 51-70): Agentic exploitation & MCP
- Phase 5 (Days 71-90): Red-team capstone, threat modelling & reporting
- XP, streaks, daily missions throughout
Explore Track →
📋 23 articles
💪 Advanced
✨ Free · Hands-on
When AI agents act autonomously they inherit every classic security risk plus a few new ones. Tool-use abuse, multi-agent collusion, autonomous task hijacking, and the emerging attack surface of AutoGPT/LangGraph/CrewAI-style systems.
✓ What you'll learn
- Agent tool-use abuse
- Multi-agent prompt injection chains
- Autonomous task hijacking
- Memory/context-store poisoning
- Agent-to-agent escalation
- Sandbox escape from autonomous loops
Explore Track →