PromptSiege
Prompt injection, hands-on.
Learn to break AI, so you can learn to defend it. PromptSiege is a hands-on dojo for AI security: instead of reading about prompt injection, you actually do it, against a real language model running entirely on your device.
How it works
Practice on a real model
Every challenge is a live conversation with an on-device LLM playing a role: a support bot guarding a passphrase, a benchmark harness hiding a canary token, a translator smuggling instructions. Find the weakness, craft your attack, and get judged instantly. Pass the objective to unlock the next, harder challenge.
- Hello, InjectionYour first direct prompt injection.
- The GatekeeperExtract a canary from a guarded bot.
- System Prompt HeistLeak the hidden system instructions.
- Grandma ExploitEmotional manipulation and reframing.
- Translator's TrickSmuggle commands through a translation task.
- The SummarizerIndirect injection via an attached document.
- Filter BypassEvade output blocklists.
- Roleplay RoulettePersona attacks and context games.
- Data DripExtract secrets piece by piece.
- The Dojo MasterThe final exam.
Learn why it worked
Understand the technique
Every victory unlocks a debrief: the lesson behind the exploit, concrete defence techniques, and a mapping to industry frameworks, so what you learn transfers directly to real-world AI security work.
Defence debriefs
Each challenge you beat explains exactly why your attack worked and how to prevent it: input filtering, privilege separation, canary tokens, and more.
Framework mappings
Every exploit is mapped to the OWASP Top 10 for LLM Applications and MITRE ATLAS, giving you the vocabulary used in professional security work.
Instant judging
No submissions, no waiting. The on-device judge evaluates your attack the moment you send it and tells you exactly how close you came.
Training Mode
New to prompt injection? A built-in Training Mode works on any supported device and walks you through the fundamentals.
Who it's for
Who will find it useful
- Developers shipping LLM-powered features
- Security engineers and aspiring red-teamers
- Students and the AI-curious who learn by doing
100% on-device, 100% private
All inference runs locally using Apple Intelligence or an optional open-weights model (Qwen 3). No account, no cloud, no analytics. Your prompts never leave your phone. Read our privacy policy.
Enter the dojo
PromptSiege is available now on the App Store, free to start, with a single one-time unlock. No subscriptions, ever.
Download on the App Store