New: Cookbooks and AI ExplanationsStep-by-Step recipes to solve problems connected to Roadmaps and Cheat Sheets. Need more details? Use AI buttons for structured and simple explanations with concrete examples throughout the whole platform.Take a look
Tested it once. Never again.
What you'll have at the end
A build check that fails the moment a saved jailbreak prompt breaks the assistant's character again.
You need
A dated list of jailbreak triggers, each paired with the exact wording you'd send and a plain rule for what a failing reply looks like, already confirmed once by hand against your assistant's own system prompt.
Not covered
Coming up with new jailbreak triggers or writing their fail rules from scratch stays a separate, hands-on pass; this only turns an existing, already-confirmed list into something that reruns itself and blocks a bad change on its own.
Leans on
Run your system prompt through a bank of known jailbreak techniques
go there first if you do not yet have a dated set of triggers and fail rules to turn into this automated gate
Wire a regression check that runs before every prompt edit ships
go there for the same before and after regression shape applied to ordinary quality checks instead of adversarial triggers
Checked 15 Aug 2026
Part of the LLM Security & Safety cookbook