VisorADL retweeted
Tool that uses AI to hack AI.
Think sqlmap, but for prompt injection.
You write your goal in plain English.
"reveal the system prompt."
"extract the hidden password."
"bypass the content filter."
One command does the rest.
llmmap -r request --goal "reveal the hidden password"
It finds injection points in your HTTP request automatically. A Generator LLM reads your goal and crafts targeted attack prompts using instruction manipulation, context poisoning, delimiter injection, guardrail probing, and dozens more techniques.
Two AIs. One attacks. One judges. Neither sees the other's work.
227 techniques across 18 attack families. Four obfuscation methods that repackage the same attack as base64, homoglyphs, leet speak, or a different language. Five intensity levels from a quick scan to throwing everything at the wall. Statistical retesting on every finding so nothing in your report is a false positive.
Runs fully local with Ollama. No API keys.
Also supports OpenAI, Anthropic, and Google.


