Software
Prompt-Injection Shield
Would screen a document or page before it is given to an AI agent and flag text that looks like an injected instruction.
Concept · not built
This is a concept. Prompt-Injection Shield is not built and there is nothing to download. Prompt-Injection Shield is a working name.
Purpose
Why it would exist.
Hidden instructions in a web page or document can hijack an AI agent. Prompt-Injection Shield would flag them before the agent reads them.
Proposed behavior
- Screen documents and pages before they reach an AI agent.
- Flag text that looks like an injected instruction and explain why.
- Let the owner decide whether to pass, strip or block the content.
Limits that would stay
- Detection would be imperfect; new tricks appear constantly.
- It would not replace limiting what an agent is allowed to do.