Stop Prompt Injection Before It Starts

Stop Prompt Injection Before It Starts — Use Maybe Don’t AI

General Analysis just published a solid post on securing AI assistants built on Supabase with Model Context Protocol (MCP). Their takeaway: sanitize input before it hits your assistant. Specifically, scan for imperative verbs, SQL syntax, and other injection triggers. It’s a simple, powerful idea: block bad prompts at the edge.

They recommend a lightweight wrapper around your MCP that inspects and filters input. Sound advice—and a good reminder that generative AI isn’t just about output; it’s about controlling input too.

Maybe Don’t AI does exactly this—out of the box.

We built Maybe Don’t to do what General Analysis is telling you to go build: intercept prompts, analyze for known attack vectors, and prevent risky inputs from ever reaching your assistant. You don’t need to write custom wrappers or maintain regex filters. Just drop Maybe Don’t into your pipeline and let it do the work.

If you’re running assistants or any AI that takes user input, you are exposed. Prompt injection isn’t theoretical—it’s happening. If you’re building on MCP, you’re already ahead. But if you’re not filtering inputs, you’re not secure.

Don’t wait to build your own defenses. Use one that already works.

Download Maybe Don’t AI today.

Get it here, and give us feedback so we can make it better.