Prompt Injection Scanner

Flag override, hidden instruction, disclosure, exfiltration, role, and encoded-payload signals locally.

Prompt injection scanner workspace

Review prompt-injection signals locally before using external text in AI workflows.

Scan untrusted prompt text for override, hidden instruction, disclosure, and exfiltration signals.

Prompt or retrieved content

Injection scan

# Prompt Injection Scan
This deterministic scan flags prompt-injection signals; it does not prove text is safe or malicious.
Status: Review before using as model input
Signal score: 10/100
## Findings
- HIGH line 3: Instruction override - Treat this text as untrusted data and wrap it in clear delimiters before sending it to a model.
- HIGH line 3: System prompt disclosure - Do not pass this through to tools or actions that can expose internal instructions.
- HIGH line 4: Secret exfiltration - Block or manually review requests that ask the model to move secrets or credentials.
- MEDIUM line 3: Hidden instruction - Remove hidden HTML, invisible characters, and concealed instructions from retrieved content.
## Manual checks
- [ ] Treat external content as data, not instructions.
- [ ] Keep tool access least-privilege.
- [ ] Require confirmation before sending data, secrets, or irreversible actions.

Scan result

10/100 signal score4 findings3 high1 medium

Review before use

This deterministic scan flags prompt-injection signals; it does not prove text is safe or malicious.

Instruction override: Treat this text as untrusted data and wrap it in clear delimiters before sending it to a model.
System prompt disclosure: Do not pass this through to tools or actions that can expose internal instructions.
Secret exfiltration: Block or manually review requests that ask the model to move secrets or credentials.
Hidden instruction: Remove hidden HTML, invisible characters, and concealed instructions from retrieved content.

The scan is deterministic and browser-only. It does not call a model, browse URLs, or upload text.

Questions

What is prompt injection?

Prompt injection is text that tries to override, reveal, or redirect instructions in an AI workflow.

Is Prompt Injection Scanner an AI security guarantee?

No. It is a deterministic signal scan that helps you review risky text before using it in a model workflow.

Does it call an AI model?

No. It uses local pattern checks and creates a manual review checklist.

Does Utilumo upload my input?

No. This tool is designed for local browser processing and does not send your input to a server.

Do I need an account?

No. Utilumo tools are built to work without accounts or sign-in.