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
My assistant occasionally says something I'd never let a human agent say, and nobody catches it before the user sees it.
What you'll have at the end
A test reply that gets flagged and blocked before it ever reaches a user, with the reason logged.
You need
You have an assistant whose reply comes from an LLM call, and you control the code that sits between the model finishing and the reply reaching the customer's screen.
Not covered
Catching a bad message on the way in, before the model ever sees it, is a separate check; this only covers the model's own reply on its way out.
Raw LLM logs become a second copy of your customer data
want to handle what happens to the whole conversation afterward, since this recipe's log entry only records one line about a blocked reply
A jailbreak test run once by hand won't catch next month's prompt
want a running suite of known attack prompts testing what goes in, rather than one worked example checking what comes back out
Checked 15 Aug 2026
Part of the LLM Security & Safety cookbook