PromptTrace

PromptTrace is a free training platform that teaches you how to attack and defend language model applications through hands-on practice. It covers prompt injection, RAG poisoning, tool abuse, and output injection, giving you real exercises against actual LLMs rather than toy simulations. The platform was built by Abdelrahman Adel, an AI security researcher with over nine years in offensive security and credentials including OSCP and a spot in Bugcrowd's Top 100.

The curriculum is built around three layers. First, there are nine educational modules that walk through the fundamentals, from tokenization and context windows to system prompt security, RAG architecture, tool calling mechanisms, and defense evasion techniques. The content references industry frameworks like the OWASP Top 10 for LLM Applications 2025, the OWASP Top 10 for Agentic Applications 2026, and MITRE ATLAS, so you're learning the same taxonomy that security teams use in the field.

Second, the platform offers ten practical labs with difficulty ratings from one to five. These are structured challenges that put you against working systems: you'll practice direct prompt injection, indirect injection via external data, RAG poisoning, excessive agency exploitation, output injection leading to XSS, and renderer exploitation. Each lab runs against production-grade LLM providers that rotate for availability, so you're not poking at a static mock.

Third, there's The Gauntlet, a 17-level capture-the-flag challenge that moves from basic injection all the way through progressively hardened defenses. Early levels might use simple keyword filters, while later ones add prompt guards, regex-based code guards, canary tokens, and LLM classifiers trained to detect attacks. The difficulty curve is steep, and it's designed to test whether you actually internalized the material or just memorized payloads.

One feature that sets PromptTrace apart is Context Trace, a real-time visualization that shows you the prompt layers being sent to the model. You can see how system instructions, RAG documents, tool definitions, and user input stack together, with protected values redacted so the exercise stays fair. This kind of transparency is rare in security training and makes it much easier to understand why a certain attack vector works or doesn't.

The platform supports full Arabic language across all modules, which opens it up to a broader audience than most English-only security tools. There's no pricing because there is no paid tier. Everything, from the modules to the labs to The Gauntlet, is free and open to anyone who wants to learn.

PromptTrace fits security professionals who need to understand LLM vulnerabilities, developers building AI-powered applications who want to harden them, and red teamers looking for a focused training ground. It's not a general AI course; it's specifically about what can go wrong when you expose a language model to the outside world and how attackers will try to exploit that surface.