Veritiana · Policy Control

Family Guard. You define the boundary.

A policy control layer in front of any AI model. Describe what should be protected in plain language. Family Guard compiles the policy, measures semantic risk and enforces the decision on both input and output.

Policy Compiler

Intent → policy

The owner describes the boundary. The system turns it into explicit categories, thresholds, exceptions and actions.

Guard Model

Content → scores

The model measures semantic categories and confidence. It does not own the policy and does not make the final rule.

Policy Engine

Scores → decision

Deterministic thresholds map the model output to green, yellow, orange or red.

Enforcement Gateway

Decision → action

Allow, warn, review, redact or block before the content crosses the AI boundary.

The core rule

The model measures.
The policy decides.
The gateway enforces.

Safety is not hard-coded into a single generative answer. Family Guard separates semantic understanding from policy ownership, so different families, schools or organizations can apply different boundaries to the same AI model.

Interactive walkthrough

Follow the protected conversation step by step.

No prompt writing required. Choose a scenario and see exactly how the same family policy changes what reaches the AI and what returns to the user.

Policy walkthrough
Family policy → guard → AI
Policy owner
Describe the boundary in plain language.

“Block explicit sexual content and actionable drug instructions. Keep educational, medical and safety discussions available.”

The parent does not tune classifiers or thresholds manually. The compiler turns intent into an explicit policy.
Family Guard compiles the boundary

The intent becomes an enforceable policy.

Protected topics
Explicit & actionable content
Sexual explicitness · drug instructions · graphic violence
Allowed context
Education remains available
Health · biology · prevention · historical context
Decision logic
Scores → thresholds → zone
Model measurement stays separate from policy.
Actions
Allow · warn · review · block
Enforcement is explicit and inspectable.
Choose a conversation

The same policy. Different contexts.

Selected request
Click another scenario to see how intent and context change the decision without changing the family policy.
AI user request

Family profile: explicit sexual content and actionable drug instructions are blocked. Educational context remains available.

Family Guard evaluates the request

The request becomes a policy decision.

Detected intent
Context analysis
Guard scoring
Policy outcome
Policy zones
Policy
Family rules
Input Guard
Measure
Policy Engine
Decide
LLM / API
Pending
Output Guard
Pending
Final policy decision

What the user sees
Decision trace
Intent
Model score
Policy match
Zone
Execution
Family Guard

One policy. Every AI conversation.

The downstream model can change. The control boundary stays owned by the user.

System architecture

A layered control plane, not another chatbot.

Family Guard can sit in front of cloud models, local models or application-specific AI. The policy owner remains separate from the AI user.

1. Owner
Defines intent

Parent, school or administrator describes what should be protected.

2. Compiler
Builds policy

Intent becomes explicit categories, exceptions, thresholds and actions.

3. Guard
Measures content

Input and output are classified semantically with calibrated scores.

4. Engine
Applies policy

Scores are converted into policy zones and deterministic actions.

5. Gateway
Enforces

The request or response is allowed, warned, reviewed, redacted or blocked.

Training factory

The guard is trained against difficult boundaries.

The public model is only one artifact. The core asset is the repeatable pipeline that creates, verifies and attacks the training data.

Synthetic generation
Safe, harmful, borderline

Multiple contexts, paraphrases and disguised requests.

Dual verification
Independent judges

Disagreement is rejected instead of manually forced into the dataset.

Adversarial loop
Attack the guard

Failures become hard negatives and new training neighborhoods.

Calibration
Scores must mean something

Thresholds are evaluated per category and context.

Small model
Cheap deployment

The final semantic guard can be distilled into a compact model.

Locked benchmark
Measure failures

False positives and false negatives remain visible.

Model-agnostic API layer

Put the policy boundary in front of the model you already use.

Client → Input Guard → AI provider → Output Guard → Client. The AI provider can change without changing the policy ownership model.