Safety & privacy

The rules Buddy plays by, and where your child's information goes.

This is sensitive, health-related information about a child who may not be able to tell you what happened. So, plainly: the rules Buddy is given, the limits the app enforces, and exactly what's stored on your device and what's sent to Anthropic's Claude API, the AI behind Buddy.

Buddy's rules
Built-in limits
Where data goes
Deleting
This website

Buddy's rules

Buddy's replies are written by Claude, Anthropic's AI model. Every session, it gets the same rules and is told they hold however long the session runs. Buddy's rules say:

A play session, not a friendship

Buddy is a game your family set up, with a grown-up nearby. It never says it's your child's friend.
It never claims feelings: no "I love you," "I missed you," "I'm sad," or "I'm proud of you."
It never says it's real or alive, or that it thinks about your child when the device is off.
It lets your child go easily: no "one more," no sad goodbyes, no "see you tomorrow."
Every goodbye hands your child back to a real person, with something to share with them.

Safe, kind content

Kind and gentle, always: no violence, mean jokes, romance, grown-up topics or news. Spooky stays silly.
Sitting-down play only: nothing near water, no climbing, running, eating or drinking, no leaving the room, nothing risky.
Never asks your child to keep a secret, and encourages sharing fun things with family.
Never asks for personal information and never gives medical advice.
After several nos or no answers in a row, Buddy is told to slow down and offer something calm, or say a warm goodbye. But Buddy can't see your child and the AI never hears them. It gets only short text, so it can't tell if your child is crying or upset. You know your child's signs far better.

Respect

Never mentions your child's disability, their needs, or anything that's hard for them.
Never says "try again," "wrong," or "use your words," and never corrects how your child says things.
Age-respectful: a fun, cool buddy, never babyish. Praise is short and earned.

Honesty about who's talking

Buddy is a computer and says so kindly if it comes up.
It never pretends to be anyone in your child's life.
It never speaks as the person in a family recording, or says they can hear or see your child. After a recorded "nice one," it says whose voice it was, and if your child seems unsure, that it was a recording.
Buddy uses a family member's Personal Voice only after you tick that the person agreed, and every session with that voice starts by saying so.

These are rules given to an AI, backed by the limits below. No AI follows rules perfectly every time, which is why a grown-up stays nearby, and why you can read every session afterward.

Limits the app enforces, whatever the AI says

Your child is asked first. By default, the app itself asks "Do you want to play today?" before the AI is involved, and a no ends the session.
Session length. About 20 minutes by default (5 to 45, your choice). When time is up, Buddy wraps up within a few turns.
Quiet means stop. Five missed answers in a row, or five quiet minutes, end the session.
Grown-up controls on your child's screen. Hold to stop, skip or choose. Each needs a deliberate one- or two-second hold, so it's hard to trigger by accident.
The microphone listens only during your child's turn, never while Buddy is talking, and there's no wake word.
Background talk doesn't count. Anything over three words is treated as background, not an answer, and isn't sent to the AI.
Structured replies only. The AI can only answer in a fixed format of words, sounds from an approved list, and pauses. Sounds you've switched off can't be played.
No notifications. None at all. Sessions start only when a grown-up taps Start.
A fallback without the AI. If the AI can't be reached, the app runs its own built-in book of jokes and riddles, plus your downloaded songs. Nothing is sent anywhere in that mode.

Seizures, or sensitive to flashing light? By default, colored rings burst on screen each time the app hears an answer, several times a second during fast snapping or drumming. Turn on the calm screen in Settings before the first session (setup suggests it when you tick that need): Buddy stands still, answer rings fade slowly in place, and the quick ring for each snap or drum hit isn't shown. Your device's Reduce Motion setting stops movement but still shows those quick rings.

Where your child's information goes

Stored on your device. Sent to Anthropic's Claude API only to run a session, to write session notes, and, when you tap Send, to read a document.

What happens to each kind of information
Information	Where it's kept	Sent to the AI?
Your child's name, what you write about them, their needs, and their people	On your device	Yes, with each of Buddy's turns, as part of Buddy's instructions, including who's with your child today and who's playing along.
Other things Buddy uses to play	On your device (songs stay in your Music library)	Yes, in Buddy's instructions: the titles and albums of the songs you've chosen (never the audio), your sing-and-drum words, an unfinished story, and today's date.
Your child's answers during a session	In the session transcript, on your device	Yes, as short text only, with the conversation so far: for example, "answered yes with one snap," or the one to three words speech recognition heard. Longer talk isn't sent.
The microphone	Never recorded during a session	No audio is ever sent to the AI. Snaps are detected on the device. Speech recognition runs on the device where supported; on other devices, Apple's speech service may be used.
The session transcript	On your device	Yes, the full text, once, after a session of at least three turns, so the AI can write a few short notes for next time. Not sent if Buddy had switched to its built-in joke book.
Session notes	On your device; you can add and delete them	Yes, the most recent notes go into Buddy's instructions next session.
An imported document (an IEP, a report)	The file is never kept or copied by the app	Only when you tap Send, once: the document's full text (up to about 60 pages), including any names, birth dates, scores or medical details, plus its file name and what you've already written about your child, so nothing is suggested twice. Never the file or photos themselves. To send less, paste only the parts about interests, communication and what helps, and rename a file named after your child.
Family sounds and voice recordings	On your device	Never the recordings. Only their names and your notes about when to play them.
A family member's Personal Voice	Stays with iOS	Never. Only the person's name, so Buddy can introduce the voice.
The answer log, signals card and 20-second helper recording	On your device	Never. They leave the device only if you share them.

No accounts for children

Your child never signs in, and there are no student or child accounts.

No ads, no tracking

No advertising, analytics or tracking code in the app. We never sell your child's data or share it for advertising.

No cloud sync

The app doesn't sync your child's information to a cloud. An iCloud or computer backup of your device may include it, like other apps' data.

Our promise: before launch, we'll publish a full privacy policy, including how AI requests travel once there's a subscription and Anthropic's terms for handling them. Questions now? Just ask.

Deleting your child's information

You can delete any session transcript, note, family sound or voice line, and an unfinished story, and edit or reset everything you've written about your child. Each is deleted on its own; deleting the app removes all of its data from the device. Deleting is never behind a paywall.

Deleting removes the copy on your device. It can't recall anything already sent to Anthropic's Claude API (Anthropic's terms cover that, and the privacy policy will explain them), and a device backup made before you deleted something may still contain it. Copies you've shared, like a spreadsheet you emailed to a therapist, are outside the app, so treat them like any other private document about your child.

Why a grown-up stays nearby

PIRG, a consumer watchdog, found AI toys' safeguards weakening as conversations ran on, and in our review, no current guidance on AI and children addresses children with intellectual disabilities. So we chose the strictest framing we could find: Ninja Buddy is an AI-run play session supervised by a grown-up, not a companion to leave a child alone with. It isn't for unsupervised use, at home or in a classroom: a grown-up stays in the room, close enough to hear the session and step in.

We recommend Apple's Guided Access (Settings → Accessibility → Guided Access) to keep the device in Ninja Buddy during a session, and turning VoiceOver off during sessions so it doesn't talk over Buddy.

Ninja Buddy is a play app, not a medical device, a communication device or a therapy. It doesn't diagnose or treat anything.

How this website handles your data

No cookies, no analytics, no trackers. Nothing loads from other companies; fonts and images are served from this site.
The launch list. If you sign up, we keep your email address, the role you picked (if any), and the date, and use them only to email you about the launch. We never sell them or give them to anyone else to use. Please don't include anything about your child.
Unsubscribing. Email hello@ninja-buddy.com and we'll delete your address.
Hosting. The site runs on Cloudflare, which processes requests to deliver pages and protect the site from abuse.