Tools Blog

Hidden Unicode / AI Prompt Scanner

Scan to open this page

Scan text live for invisible Unicode characters — the Unicode Tags block and bidi overrides are flagged High risk (the exact categories documented in real AI agent prompt-injection research), while common zero-width, NBSP, and BOM characters are flagged Low.

How it works

  1. Paste text, a prompt, or an AI agent skill file
  2. Every hidden or unusual character is listed live below — name, code point, position, and risk level
  3. A clean copy with every flagged character removed is ready to copy instantly

Frequently asked questions

Are these characters actually dangerous?

Categories flagged High — the Unicode Tags block and bidirectional override characters — have been documented in 2026 security research (Cloud Security Alliance) as real injection vectors hidden inside AI agent skill files, tool descriptions, and MCP server metadata: invisible in every standard UI, but processed by the underlying model. Low-risk characters are usually just accidental artifacts from copy-pasting.

Does it scan as I type?

Yes — no button to click.

Is my text sent anywhere?

No, everything runs in your browser and nothing is uploaded.

#HiddenUnicodeScanner#InvisibleCharactersInTextChecker#ZeroWidthCharacterRemover#AiPromptInjectionUnicodeScanner

Related tools

Password Strength Checker Password Strength Checker Photo Privacy Checker Photo Privacy Checker

From the blog

Most invisible-character checkers treat every hidden character the same. Only some of them are the ones real attacks actually use.
security

Most invisible-character checkers treat every hidden character the same. Only some of them are the ones real attacks actually use.

We checked two invisible-character tools directly — one is genuinely live but never mentions AI agents at all, the other needs a button click despite its name. Neither separates the character categories that 2026 security research actually documented as AI agent injection vectors from the ones that are just copy-paste accidents. Ours does.

2026-08-30

Update history

  • 2026-08-30 Added: a cover image — used as the card thumbnail, social share image, and blog title image.
  • 2026-08-30 Fixed: the High/Low risk border colors on flagged characters weren't rendering (a CSS scoping issue affecting dynamically-generated content) — now shows correctly.
  • 2026-08-30 Launched: live scan for invisible/suspicious Unicode characters (no button), High/Low risk classification (Unicode Tags block and bidi overrides flagged High — the categories documented in real AI agent prompt-injection research — vs. common zero-width/NBSP/BOM flagged Low), one-click clean text.