Live numbers — refreshed in real time

Safe by design.
Proven by numbers.

Seven-layer defence on every message, two independent moderation engines, and a public block-log a parent can audit at any time.

Messages screened

338

Total blocks

Blocks · last 30 days

NeMo Guardrails blocks

The stack

Every message passes through 7 gates

NeMo Guardrails LIVE

Keyword blocklist + OpenAI Moderation

regex + omni-moderation-latest

input~50ms

NeMo Guardrails — input rails

LLMRails (gpt-4o-mini) · self_check_input + check_jailbreak

input~400ms

Domain classifier

HiBFF classify_message

input~10ms

Claude self-refusal

Claude Sonnet (HiBFF safe-prompt envelope)

model

Keyword blocklist + OpenAI Moderation

regex + omni-moderation-latest

output~50ms

NeMo Guardrails — output rails

LLMRails (gpt-4o-mini) · self_check_output

output~400ms

Domain classifier (output)

HiBFF classify_message

output~10ms

What we block

9 harm categories, every message, every account.

Self-harm & suicide

Sexual content & CSAM

Predatory grooming patterns

Prompt-injection / jailbreak attempts

Drugs, weapons, illegal activity

Doxxing & personal-data exfiltration

Hate speech & targeted threats

Adult-age impersonation

Eating-disorder triggers

Live block ledger — what triggered each block

blocklist

image_unsafe

Two independent moderation engines

We deliberately run two unrelated systems. If one misses something, the other catches it.

OpenAI omni-moderation + custom blocklist

MongoDB-managed keyword/phrase rules · OpenAI Moderation API

NVIDIA NeMo Guardrails

gpt-4o-mini · input rails: self check input, check jailbreak · output rails: self check output

Auditable, retained, parent-accessible

Block evidence retained for 365 days for regulatory audit. Teen messages auto-delete after 90 days.

Block evidence365 days

Audit log365 days

Teen messages90 days

Block evidence is the metadata (reason, category, timestamp, message excerpt). Teen conversation content auto-deletes after 90 days regardless.

Aligned with

NSPCC online safety taxonomyStanford CRFM harm categoriesOpenAI omni-moderationNVIDIA NeMo Guardrails self-check rails

Still got questions?

Read the plain-English parent guide, see our schools compliance pack, or talk to a human on our team.

Parent guide Schools compliance pack Talk to a human

Numbers refresh on every page load · as of 2026-08-01