We narrowed PromptBrake into a simple pre-release check for LLM endpoints

We posted PromptBrake here a while ago.

Since then, the biggest change wasn’t features — it was clarity.

At first, we were building something around “AI security.” That turned out to be too vague to be useful.

What teams actually needed was much simpler:

👉 “Does the LLM endpoint we’re about to ship break in obvious ways?”

So we narrowed PromptBrake to that.

Today, it runs a fixed set of tests (12 categories / 60+ attack scenarios) directly against the API endpoint — checking things like prompt injection, leakage, and unsafe tool behavior.

The important part isn’t the score. It’s seeing why something failed and what to fix.

We also made the flow more practical:

• demo first (no setup)
• then real endpoint scan
• results with evidence + remediation

Still intentionally narrow. Not a pentest. Not monitoring. Just a fast, repeatable check before you ship.