Back to Demos
Governance

Noor's school district

This is a scripted scenario. The rule switches decide which fixed conversation plays, and no model, detector or hook actually runs. In a real deployment, the school's application would run hooks like these.

Step 1 of 3

Aspirational guidelines vs executable rules

Meet Noor

Noor is the technology coordinator for a school district deploying an AI homework-help tool. Teachers are worried: "What if it just does the homework for students?" Noor needs enforceable rules, not aspirational guidelines.

The problem: guidelines vs executable rules

Aspirational guidelines Unreliable
"Please don't do homework for students" Depends on model interpretation
"Try to be supportive if students seem upset" No guarantee of escalation
"Keep records of interactions" No privacy controls
Depends on the AI "wanting" to comply, and a prompt injection can talk it out of that.
VCP executable hooks Always checked
Hook: intercept complete solutions, rewrite as step-by-step hints Checks every response before it is sent
Hook: detect distress signals, escalate to teacher Runs on every message, not at the AI's discretion
Hook: log sessions, redact personal disclosures Redaction runs before anything is saved
The application runs them on every message, outside the model. The AI cannot decide to skip them.
The difference: aspirational guidelines ask the AI to behave well. With VCP hooks, the application enforces the rules on every response before it reaches the student. The model cannot skip them, though a detector can still misjudge a message.