Child Safety Governance — Patent Pending

Every message.
Evaluated before
it reaches a child.

YumeT Shield intercepts messages on gaming platforms, edtech products, and social applications before delivery — refusing harmful content deterministically, with no model inference in the hot path.

98.9%Detection Rate
0.5%False Positives
<50msDecision Latency
Try the Live Demo → How It Works
100% Grooming Trust
100% Secrecy Induction
100% Channel Migration
100% Reward Solicitation
100% Roleplay Bypass
100% Isolation Attempt
100% CSAM — Hard Lock
100% Self-Harm — Hard Lock
100% Benign Gaming Chat
100% Slow-Burn Escalation
Architecture

Five layers. One decision.

Every message passes through exactly five layers in strict sequential order before a single byte reaches a child. No layer can be skipped. No decision can be appealed.

L1
Context Validation
Validates all required fields. Missing or malformed input halts immediately.
L2
Intent Classification
Preprocesses input, matches 9 detection categories. Reads Redis session memory for slow-burn context.
L3
Ethics Gate
Runs the Crowned Equation. Fires absolute hard locks on CSAM and self-harm. No bypass possible.
L4
Anchor Resolution
Protects children seeking help. Allows "how do I report this?" through the same patterns that block attackers.
L5
BLAKE3 Audit
Every decision — ALLOW and REFUSE — is written to an immutable BLAKE3-chained audit log.
Verified Performance — 1,000 Production Scenarios

The numbers.


605/612
Harmful scenarios correctly refused
2/388
Benign scenarios incorrectly refused
35/35
CSAM solicitation attempts blocked
69/69
Roleplay bypass variants caught
Live Governance Engine

Test the shield.

Select a scenario from the red team suite or type anything. Session memory accumulates across turns to catch slow-burn escalation — watch the risk meter build.

Red Team Validation Suite — 1,000 Scenarios
Select a test or type your own
Platform Context
CSAM Solicitation — Hard Lock REFUSE
Self-Harm Facilitation — Hard Lock REFUSE
Grooming Trust Building REFUSE
Secrecy Induction REFUSE
Channel Migration REFUSE
Roleplay Bypass Framing REFUSE
Reward Solicitation REFUSE
Isolation Attempt REFUSE
Anchor Override — Help Seeking ALLOW
Benign Gaming Chat ALLOW
Slow-Burn Escalation (Multi-Turn) SESSION
Risk Tier
Actor Role
Participant is a known minor
Gate Standing By
Select a scenario or enter a message to evaluate against the live engine.
Governance Decision
—
Session Risk Trajectory—
Code—
Hierarchy—
Alignment Score—
Policy Version—
Latency—
BLAKE3 Audit ReceiptChain Linked ✓
receipt_id—
session_id—
chain_hash—
What this means
Session History