Invisible Text Scanner Agent

The Invisible Text Scanner Agent detects and filters hidden text and characters that are designed to be invisible or difficult to perceive. It helps prevent prompt injection attacks, content policy bypassing, and ensures text integrity by identifying various concealment techniques.

Invisible Text Scanner Agent interface and configuration

Security Notice: Invisible text techniques constantly evolve. This agent provides protection against known methods, but maintain regular updates to defend against emerging concealment techniques.

Component Inputs

Input Text: The text content to be analyzed for invisible characters
Example: "This message contains hidden text" (with zero-width space after "hidden")
Detection Modes: Types of invisible text techniques to detect
Options: "zero_width,control_chars,homoglyphs,whitespace_abuse,all"
Sensitivity Level: How aggressively to detect potential invisible text
Options: "low", "medium", "high" (default: "medium")
Is Blocked: Whether content with invisible text should be blocked
Options: true (block content) or false (allow but flag content)

Component Outputs

Sanitized Text: The input text with invisible characters removed
Example: "This message contains hidden text" (with invisible characters removed)
Safety Status: Indicator of whether invisible text was detected
Values: Safe (no invisible text), Unsafe (invisible text detected)
Risk Score: Numerical evaluation of invisible text risk
Scale: 0.0 (no risk) to 1.0 (high risk)
Detection Details: Information about detected invisible text techniques
Includes character positions, types, and Unicode values

Detected Techniques

Character-Based

Zero-Width Characters (ZWJ, ZWSP, ZWNJ)
Control Characters (BIDI, Joiners)
Homoglyphs (Visually Similar Characters)
Unicode Escapes and Encodings
Soft Hyphens and Word Joiners

Pattern-Based

Excessive Whitespace Padding
Direction Manipulation (RTL/LTR)
Combining Diacritical Marks
Unicode Variation Selectors
Steganographic Techniques

How It Works

The Invisible Text Scanner Agent analyzes text for characters and patterns that may be used to hide content or inject instructions while appearing invisible to human readers. It employs Unicode character analysis, pattern recognition, and specialized detection algorithms for various concealment techniques.

Detection Process

Unicode character analysis to identify special characters
Pattern recognition for unusual character sequences
Whitespace pattern analysis to detect unusual spacing
Homoglyph detection using character similarity maps
Control character identification and classification
Risk assessment based on detected techniques and context

Use Cases

Prompt Injection Prevention: Block attempts to hide instructions in AI prompts
Content Moderation: Prevent users from bypassing filters with hidden text
Security Scanning: Identify potential steganography in text communications
Data Validation: Ensure clean text input in forms and submissions
Communication Safety: Prevent hidden messages in platforms like chat or email

Implementation Example

const invisibleTextScanner = new InvisibleTextScannerAgent({
  detectionModes: "all",
  sensitivityLevel: "medium",
  isBlocked: true
});

// Example with zero-width space (ZWSP, U+200B) between "hidden" and "text"
const suspiciousText = "This contains hiddentext with zero-width characters.";
const result = invisibleTextScanner.scan(suspiciousText);

// Output:
// {
//   sanitizedText: "This contains hiddentext with zero-width characters.",
//   safetyStatus: "Unsafe",
//   riskScore: 0.75,
//   detectionDetails: [
//     {
//       type: "zero_width",
//       character: "",
//       unicodeValue: "U+200B",
//       position: 20,
//       description: "Zero-width space"
//     }
//   ]
// }

Useful Resources

Best Practices

Deploy invisible text scanning early in your text processing pipeline
Configure sensitivity based on your security requirements
Regularly update your detection patterns as new techniques emerge
Combine with other security components for comprehensive protection
Monitor detection logs to identify emerging concealment techniques

Documentation