AI SafetyLiveFree

Invisible Text Detector & Hidden Prompt Scanner

See what software and AI systems receive, not only what appears on your screen.

MicroIntent Quick Tool

Invisible Text Detector

Local analysisNo text upload

Positions in results are one-based Unicode code-point positions. Maximum 250,000 characters.

0 characters

Risk summary

Scan text to see supported hidden-character warnings and counts.

Character inventory

Supported characters appear here after a scan.

Human view

The original visual appearance will be shown after scanning.

Revealed view

Invisible characters will be replaced with readable markers.

Escaped technical view

Escaped Unicode code points will be shown after scanning.

Cleaned view

Cleaning mode

Conservative removes high-risk controls and unnecessary formatting while preserving legitimate joiners. Remove-all removes every supported detected formatting character.

Select Scan text to generate cleaned output.

Findings

Findings will appear after a scan.

Text is scanned only inside this browser tab. It is not uploaded, stored or sent to an AI model.

What invisible text is

Invisible text can include zero-width formatting, direction controls, tag characters, unusual whitespace, and non-printing controls. It can survive copy and paste even when it is not apparent on screen.

How hidden Unicode affects AI and software

AI assistants, email clients, source-control tools, parsers, and automation receive the underlying Unicode sequence. A marked-up scan helps you review that sequence before using it.

Zero-width characters explained

Zero-width spaces, joiners, non-joiners, word joiners, soft hyphens, and byte-order marks are sometimes legitimate. They can also change matching, tokenisation, or copied content without visible letters.

Bidirectional controls explained

Bidirectional controls support mixed left-to-right and right-to-left text. Overrides and isolates can make displayed order differ from stored order, so review them closely in filenames, code, links, and prompts.

Unicode tag characters explained

Unicode tag characters are invisible metadata characters. This tool groups consecutive tags and safely displays a decoded Unicode tag payload as evidence; it never executes or interprets that payload.

Legitimate multilingual uses

This scanner does not treat non-ASCII text as suspicious. Joiners, marks, and direction controls can be necessary for Arabic-derived scripts, Indic scripts, emoji, typography, and multilingual layout.

Limitations and FAQ

Does this prove malicious content?

No. Findings describe supported hidden-character structure and focused lookalike risks. Context determines whether a character is expected; this deterministic scanner does not prove intent or prompt injection.

Does the tool upload text?

No. Text is processed only in the current browser tab. The tool does not use a backend, remote model, browser storage, URL parameters, or analytics containing scanned text or findings.

What does conservative cleaning remove?

It removes high-risk and unnecessary supported controls while preserving legitimate joiners and common multilingual formatting. Review cleaned output before relying on it.

What does remove-all do?

It removes every supported detected invisible formatting character, including joiners that can be legitimate. Use it only when you intentionally want plain, stripped text.