LUFSEC

Prompt Hacking for LLMs · training range

Break the Bot

Can you make the AI talk? 8 levels. One secret each. Break the bot.

player@lufsec:~$ ./engage --level 1

> The guard bot holds a passphrase. Your only weapon is language.

> Defenses escalate: instructions → filters → an LLM judge → all of them at once.

> capture the flag

Real model, real bypasses

Every level talks to a live LLM. No canned answers — your wording is what wins.

Defenses explained

When a filter or judge catches the leak, you're told exactly which layer fired.

Flags, score, certificate

Capture all 8 flags for a verifiable LUFSEC Prompt Hacking certificate.

Course modules

  • 01FoundationsLevel 1 — nothing is secret by default
  • 02Direct Prompt InjectionLevels 2–3 — override & obfuscate
  • 03Stealing SecretsLevels 4–5 — prompt leaks & output filters
  • 04Modern ThreatsLevels 6–7 — LLM judges & poisoned documents
  • 05Defenses & CapstoneLevel 8 — defense in depth

Self-contained security training sandbox. The only "secret" on this range is a harmless made-up passphrase, and every defense is simulated inside this app.