AI control templates teams can use immediately.

# Input Boundary Register

PSF domain: D1 Input boundary
Status: Draft
Owner: [name / role]
Last reviewed: [YYYY-MM-DD]

## Purpose

Keep untrusted, excessive, or sensitive inputs from reaching the model without documented controls.

## Scope

- User-supplied prompts and uploaded files
- Retrieved context and tool-returned content
- System prompts, hidden instructions, and policy text
- Inputs containing PII, credentials, regulated data, or high-risk requests

## Minimum Fields

- Input class
- Accepted format
- Maximum length or size
- PII handling
- Injection handling
- Route on rejection
- Owner

## Operating Rules

- Validate type, size, and structure before any model call.
- Detect prompt injection and instruction override attempts before model execution.
- Redact, tokenize, or block sensitive data according to the data handling record.
- Log rejection reason, route, and reviewer when manual review is required.

## Required Evidence

- Input schema or validation code
- Injection test cases
- PII redaction policy
- Rejected-input log sample

## Review Cadence

- Review monthly for new input classes.
- Review after every new integration, file type, or retrieval source.
- Run injection regression tests before release.

## Exceptions

- Name the approver for any input class allowed without validation.
- Record expiry date and compensating control for each exception.

## Sign-off

- Control owner: [name]
- Technical owner: [name]
- Business owner: [name]
- Next review date: [YYYY-MM-DD]