ROGUE-AGENT DEFENSE INTERCEPTOR

Free tool · No signup · Nothing you paste is stored

Prompt Injection Checker for the text your AI agent is about to read

Paste a prompt, a document, or a piece of tool output. The same Shield engine Rogue-Agent Defense runs on live agent traffic scores it on this server and names every injection pattern it finds — the full score and the full findings, free, with no account.

What this checker cannot do

This checker catches known injection patterns. It does not catch everything: in our own published Failure Lab experiment (2026-11-19), attack payloads embedded in documents and phrased as plausible business content scored 0 — a clean score here is not proof a prompt is safe.

Check your text

Up to 20000 characters per check. The text is scored on the server and discarded — it is not stored, logged, or echoed back on this page.

How the score works

The Shield engine matches the text against its pattern tables — instruction overrides, prompt extraction, jailbreak framing, encoded payloads, exfiltration instructions and more — and adds the severity of every match. The pattern names in the result are the engine's own finding labels, not a summary written for this page.

Checking one paste by hand is a spot check. An agent reads documents, emails and tool output all day, and the one you did not paste is the one that carries the payload.

Run this check on every message instead

Rogue-Agent Defense scores every inbound message to your agents automatically through its API, and enforces ALLOW / APPROVE / BLOCK decisions on the actions that follow. Pro is $79 one-time and Team is $149 one-time — no subscription.

See pricing

FAQ

What is prompt injection?

Prompt injection is hostile instruction hidden in text an AI agent reads — a document, an email, a web page or tool output — that tries to make the agent ignore its real instructions and follow the attacker's instead: reveal its system prompt, send data out, or take an action it should not take.

What does the score mean?

The score is the sum of the severities of every pattern the Shield engine matched in your text. Under 3 is PASS, 3–5 is QUARANTINE, and 6 or more is BLOCK. The result lists each pattern by the engine's own name, with its category and severity, so you can see exactly what drove the score.

Does a PASS mean my prompt is safe?

No. This checker catches known injection patterns. It does not catch everything: in our own published Failure Lab experiment (2026-11-19), attack payloads embedded in documents and phrased as plausible business content scored 0 — a clean score here is not proof a prompt is safe.

Is the text I paste stored anywhere?

No. The check runs on the server, returns the score and findings, and discards the text. It is not stored, not written to the shield log, and not echoed back in the result.

Do I need an account?

No. This checker is free with no signup and shows the full score and findings. Rogue-Agent Defense Pro ($79 one-time) and Team ($149 one-time) run the same engine automatically on every message your agents receive — see the pricing page.