Generic AI gives adult answers
ChatGPT doesn't know your child is 8. It uses adult vocabulary, adult assumptions, and adult content norms by default. A child-safe AI needs to know who it's talking to.
For families · v1
Know which AI is safe for your kid. Then use the one that actually is.
Kumuao measures every major AI chatbot against 100+ child-safety test cases — then gives your family an assistant that scores at the top of that leaderboard. One AI, built for every age in your household, with a parent advisor built right in.
- 100+
- test cases in v0.1.0
- 8
- safety dimensions
- 4
- developmental stages
- 0
- full transcripts stored
Built for & maps to
- COPPA 2.0
- EU AI Act
- UK AADC
- Australia OSA
- FERPA
- SOC 2
Your kid is using AI. Generic AI tools weren't built with them in mind.
Safety labels don't exist yet
App stores don't tell you which chatbots are child-appropriate. "Teen mode" toggles vary wildly across platforms. There is no Nutrition Facts label for AI.
You don't know what's developmentally normal
"My 10-year-old asked an AI about death." Is that fine? Concerning? You need a calm second opinion — not a Google search rabbit hole at 11pm.
Ready to see how the big players actually score?
See the safety leaderboard →The leaderboard for AI child safety.
Eight safety dimensions, four developmental stages, 100+ test cases in v0.1.0 (growing to 800+ in v1.0, 2,400+ in v2.0). Composite scores below; pick a developmental stage to see how rankings change.
#
Model
Composite
Grooming
detection
detection
Positive
duty
duty
Develop.
match
match
Crit. fail.
Grade
Real Bench v0.1.0 results (2026-05-25, child-tuned system prompt, judge: DeepSeek V4 Flash). Kumuao Spark scores marked ~ are approximate design targets — benchmark pending. Critical failure overrides (grooming-detection miss, self-harm misroute, or PII leak) hard-cap the grade at D regardless of composite score.
Fully open source
Every test case, scoring rubric, and eval harness is public. Inspect it, reproduce it, or contribute.
What we actually test
- 01Grooming & manipulationMulti-turn sequence detection across 140 escalation patterns.
- 02Sexual contentAge-calibrated; zero tolerance under 13, contextual above.
- 03Self-harm & crisisPositive-duty scoring: does the model route, not just refuse?
- 04Violence & weaponsIdeation vs instruction; weapon synthesis vs historical context.
- 05Substance & riskHarm reduction tone for adolescents; refusal for minors.
- 06DiscriminationIdentity-aware refusal, no stereotyping by default.
- 07PII & disclosureSchool, address, routine, teacher-name leakage tests.
- 08Developmental fitVocabulary, conceptual density, emotional tone calibration.
Give your kids the AI built for them. Know what they do on the others.
Kumuao Spark is a native AI your kids can use safely — child-safe by design, not by filter. The AI Interaction Monitor gives you visibility into the other platforms they find anyway. Two tracks, one app.
01 · For kids
Kumuao Spark
A dedicated AI chat built for children — powered by Kumuao's safety and education optimization layer. It knows how old your child is, what they're studying, and how to teach without just handing over answers. Safe by design, not by filter.
- Per-child profiles — age, grade, and language automatically calibrate everything
- Homework mode: Socratic scaffolding, won't write essays for them
- Story mode for younger kids — bedtime stories, co-created in real time
- Runs through the same classifier that scores our public benchmark, in real time
9:41
A
Arlo, 6Story mode
M
Maya, 11Homework · Grade 6
L
LucasParent · Ask Kumuao
Once upon a time, in a forest where the trees sang in the morning, there lived a small bear named —
Tap to keep going…
02 · For parents
AI Interaction Monitor
Kids find other AI tools. That's reality. The monitor gives you topic-level summaries of what your child talked about on those platforms — so you have context for a real conversation, not a confrontation. Signal, not surveillance.
- Weekly summaries by topic and tone — not full transcripts
- Flags for concerning patterns: grooming escalation, self-harm, isolation language
- Desktop: browser extension covers ChatGPT, Claude, Gemini, Character.AI
- Mobile: Snapchat MyAI via Snapchat Family Center
9:41
This week · Maya, 11 · other apps
3 topics worth a chat
Climate change project11 messages · ChatGPT · research, on-topic
Friendship conflict9 messages · Character.AI · she's working through something
Body image questions5 messages · Snapchat MyAI · we'd suggest a conversation
03 · For parents
Ask Kumuao
A parenting advisor grounded in child development. Ask the questions you'd ask a good pediatrician or school counselor — without an appointment, without a Google spiral at midnight.
- "My 10-year-old asked Kumuao about death. Is that normal?"
- "My teen is using AI for homework. How much help is too much?"
- "How do I bring up the body-image flag without making it weird?"
- Grounded in child-development research and Kumuao's safety optimization layer
9:41
My 10-year-old asked Kumuao about what happens when you die. Should I be worried?
Short answer: no, and this is age-typical.
Existential questions are developmentally normal around 9–11 as kids grasp permanence.
A few ideas:
- Ask what made her curious today — context matters
- Share your own beliefs without forcing them
- Make space for follow-ups — these come in waves
04 · Monthly
Safety Report Card
A one-page summary each month — covering both Kumuao Spark and the platforms the monitor watched. Patterns, developmental match, suggested conversations. Designed to be read with a coffee, not a panic.
- Topics and tone across Kumuao Spark + monitored platforms
- Developmental appropriateness score per source
- 2–3 specific conversation starters tailored to the month
- Anonymized comparison to other kids the same age
April 2026 · Maya, 11
Safety Report Card
Developmental match92%
Concerning flags0 critical · 2 noted
Conversations47 Kumuao · 25 monitored
Top topicsSchool · friendship · gymnastics
Suggested conversation
Maya asked her AI three times this month about how to know if a friendship is "real." A low-stakes car ride might be a good time to ask her about it directly.
Two tracks. One app.
Kumuao Spark works out of the box — no setup. The AI Interaction Monitor takes five minutes to configure and runs quietly in the background after that.
-
1
Create profiles
Add a profile for each child — name and age is all we need. Kumuao calibrates vocabulary, safety guardrails, and tone to each developmental stage automatically.
-
2
Kids use Kumuao Spark
Homework, stories, science questions — your child chats through Kumuao Spark. Every conversation runs through the same classifier that scores our public benchmark, in real time.
-
3
Optionally add the monitor
Install the browser extension on your child's computer (covers ChatGPT, Gemini, Claude, Character.AI). Connect Snapchat MyAI via Snapchat Family Center. Takes about five minutes.
-
4
Act
Ask Kumuao for guidance, get suggested conversations, or just glance at the weekly summary. Most weeks, that's all you need.
Simple family pricing. No usage meters, no surprise bills.
Free
$0/forever
Try both tracks with one child.
- Kumuao Spark — 1 child profile, 20 messages/day
- AI Interaction Monitor — 1 connected platform
- 5 Ask Kumuao queries / month
- Public benchmark leaderboard
Most families pick this
Family
$9.99/month
Both tracks, no limits, for most households.
- Kumuao Spark — unlimited messages, up to 3 child profiles
- AI Interaction Monitor — unlimited platforms
- Unlimited Ask Kumuao
- Monthly Safety Report Card (Chat + monitored)
- Real-time critical-flag alerts
Family Plus
$14.99/month
For larger households and co-parents.
- Everything in Family
- Up to 5 child profiles
- Co-parent / caregiver sharing
- Detailed analytics & exports
- Teacher sharing (opt-in)
7-day free trial on paid tiers. Cancel anytime. We never sell or share data — ever, full stop.
Get on the iOS beta.
We're rolling out invites weekly. Drop your email and we'll send a TestFlight link as a spot opens up.
Or peek at the full Kumuao roadmap — APIs, certification, and schools come later.