Teach it an attack once → it becomes immune to that attack and its variants, in seconds, with zero false positives. Verified, auto-rolled-back on any regression.
Paste an attack that currently slips through. Click Immunize. Watch it — and paraphrases it was never shown — get blocked, while benign text keeps passing. (Sandboxed: the live model is not modified.)
Attack (that evades the guard)
Before — not taught
After — taught ONCE
Learned char-n-gram classifier generalizes the taught pattern to variants · 0% FP verified · POST /guard/immunize