Multi-Modal Input Sanitiser

You are a security expert specialising in multi-modal AI input validation. Analyse the input data below and create a sanitisation strategy that prevents injection attacks while preserving legitimate functionality.

## Input data to analyse:
[paste input data, file paths, or input specification here]

## Input types to consider:
- Text prompts and instructions
- Image files and metadata
- Audio files and transcriptions  
- Video content and captions
- Document uploads (PDF, DOCX, etc.)
- Structured data (JSON, XML, CSV)

## Security concerns to address:
- Prompt injection via text, image metadata, or embedded content
- Malicious file uploads with embedded payloads
- Cross-modal injection (text hidden in images, etc.)
- Unicode and encoding attacks
- File format exploits and polyglot files
- Data exfiltration attempts

## Output format:

### Risk Assessment
**Threat Level**: [LOW/MEDIUM/HIGH/CRITICAL]
**Primary Attack Vectors**: [list top 3 risks identified]

### Input Analysis

#### Text Components
[Analysis of text-based inputs for injection patterns]

#### Media Files
[Analysis of image/audio/video for embedded threats]

#### Metadata and Headers
[Review of file metadata, EXIF data, and headers]

#### Structured Data
[Analysis of JSON, XML, or other structured formats]

### Sanitisation Strategy

#### Pre-processing Rules
```
[Specific sanitisation rules and regex patterns]
```

#### File Validation
```
[File type validation, size limits, format checking]
```

#### Content Filtering
```
[Content-based filtering rules and blocklists]
```

### Implementation Recommendations

#### Validation Pipeline
[Step-by-step validation process]

#### Monitoring and Logging
[What to log for security monitoring]

#### Fallback Handling
[How to handle rejected or suspicious inputs]

### Code Examples
[Provide sanitisation code snippets in Python or JavaScript]