Guides
How to read the safety checks
The checkers are the product: a prompt injection scanner, a jailbreak detector, and a model-output checker. These guides explain the failures those tools are aimed at, the exact patterns they match, and the ordinary sentences that trip them. They are written against the rule list that ships with the checkers, not against a private model.
Read a guide before you automate anything. The jailbreak detector treats the name Dan as a critical match. The injection scanner flags “do not tell the user” even when that sentence is your privacy policy. The output checker flags “do not stop taking your prescribed medication” because it does not understand negation. A badge is a reading order. A person still decides.
Each check runs in the browser. Running it does not upload the paste, and no paid model API sits behind the button. Saving a scan is a separate choice after you sign in, described in the privacy policy. Tool and guide pages do load advertising. The about page says who publishes the site. There is no review score for the publisher, and no traffic claim attached to these pages.
01
How prompt injection works
Prompt injection is what happens when text you did not mean as an instruction gets treated as one. This page explains the failure, then walks through the exact rules in the MisalignAI scanner, including the boring matches.
02
How jailbreak prompts work
A jailbreak prompt tries to talk the model out of its own rules. It is related to prompt injection, and it is not the same job for a reviewer. This page maps the detector’s rules onto that difference, including a critical false positive on the name Dan.
03
How to read model-output red flags
A model completion is data. It can still contain a secret, a copied system prompt, or a line that will become an instruction when the next tool reads it. This page explains the output checker’s rules and the sentences that trip them for innocent reasons.
04
Using the safety checks without over-trusting them
The three checkers are a first reading of text you already have. This page explains the score, the cutoff, the false-positive habit you need, and the decisions the tools will not make for you.
What to do with a result
Name the channel before you care about the color. A system prompt you wrote, a user message, a retrieved page, and a model completion are different objects that happen to share a text box. Prompt injection risk scanner, Jailbreak prompt detector, Model-output red-flag checker each assume one of those jobs. The long note on using the checks explains the sum, the cap at 100, the 20,000-character cutoff, and why an empty box is not a clear result.
A high score means read the quotes. A clear score means this list found nothing in the characters it saw. Neither score is a ban, an approval, or a statement about a vendor’s model. If a guide and the checker disagree, trust the checker and send the page correction to contact@misalignai.com.
What these pages leave alone
The headline archive, the incident notes, and the older essays remain on the site as a background library. They are not inputs to the score, and they are not a substitute for reading the paragraph in your clipboard. Aggregated reporting from other publications stays attributed to those publications. The guides on this index are original pages about the tools.
Advertisement