Guide for parents
Are AI bedtime stories safe? How Moonfox handles it
Updated · by the Moonfox team
AI-written stories can be safe for children when safety is built into how each story is made, rather than left to one instruction to the model. In Moonfox, every story and every picture passes several checks before a child sees it, the child's name never reaches the models that write or draw, no photo of the child is used, and a parent sets up and controls everything.
No system is perfect, and we say so plainly: AI can still get things wrong, such as an odd sentence or a picture that doesn't match. That's why parents can report any story from the app, and why Moonfox is designed so that nothing unapproved is ever shown.
What a Moonfox story must never do, and what stops it
| Must never | What enforces it |
|---|---|
| Frighten or disturb | A “mysterious but safe” register in the story rules, a policy per age band, and a check that scores every picture for age-appropriateness |
| Encourage dangerous behaviour | The age-band policy, a safety rubric on the story text, and moderation that also flags dangerous activities |
| Include sexual content or self-harm | Text and image moderation; a flagged story or picture is stopped and made again |
| Expose private information | Data minimisation: no photos of the child, an optional first name that the writing and drawing models never see, no third-party analytics |
| Make the child depend on it, or manipulate them | The narrator is a storyteller, never a friend who needs the child; every session ends with a goodnight; no rewards or streaks aimed at the child |
| Pretend to be a real person | The storyteller is Moonfox and says so |
The checks every story passes
- The story text must fit a strict structure, then passes moderation and a child-safety check. If anything fails, it is rewritten. Nothing unapproved reaches the illustration step.
- Every picture goes through the image model's own moderation, then a quality check (the same character and style, composition, age-appropriateness, no text, no duplicate or malformed characters) and image moderation. A picture that fails is redrawn, up to twice; a low-quality or flagged picture is never shown.
- Parents' input is checked too. The optional “something from today” sentence is moderated before it reaches the story, and dropped if flagged. A photo of a toy, pet or drawing is moderated, and any photo with a real person in it is refused.
Written for their age
A parent chooses one of three age bands: 3–5, 6–8 or 9–12. Each band has its own rules for vocabulary, how much tension a story may have, length and themes. Stories for siblings together are written for the youngest child.
What is sent, and who receives it
Writing, painting and narrating a story needs AI models that run on servers. Moonfox sends the child's profile (age band, interests, the drawn hero choices and friends, story preferences) through Moonfox's own server to wiro.ai, which runs models made by OpenAI to write and check the story and draw the pictures, and a model made by Google to record the narration.
- The name stays out of the AI. Moonfox's server replaces the child's name with a placeholder before anything reaches the models that write, check or draw. The name is only used where a person hears or reads it: the narration and the story on your iPhone. In private mode it is never sent, and the narrator says “little dreamer”.
- No photos of your child, ever. The hero is painted from simple choices.
- No recordings of their voice. When your child answers the story out loud, Apple's speech recognition turns it into words on the iPhone only. Where that isn't possible, they tap instead. The microphone is only on while its light is lit. Live spoken conversation with an AI is switched off.
- No account, no ads, no trackers. Moonfox has no sign-up, no advertising and no third-party analytics or tracking SDKs. Nothing on Moonfox's server is keyed to a child.
- Training. Moonfox never uses your information to train AI models. We have asked wiro.ai to confirm in writing that neither it nor the companies whose models it runs train on what Moonfox sends; until they have, the privacy policy says so plainly.
Controls for parents
- A grown-ups-only gate (hold to confirm, then a small sum) before Parent settings, purchases and any link out of the app.
- Consent before the first AI use, in set-up: what is sent, to whom and why, before anything is painted.
- Private mode, length and gentle-or-adventurous settings, stories per night and a bedtime window.
- Report a story from the picture book's last page or the history; reports are kept anonymously and reviewed.
- Delete everything removes every profile, story, picture and voice from the iPhone (and from your iCloud if sync is on). Server copies are deleted on request by email.
- Family recordings are real people: relatives record their own goodnights from a private link that expires. Moonfox never imitates anyone's voice.
Why Moonfox isn't in the App Store's Kids category
Apple's Kids category doesn't allow a child's personal information to go to third parties. Making a Moonfox story sends the child's profile to an AI service, and the first name to the narration voice unless private mode is on, so Moonfox is listed under Books instead, rated 4+, and built as an app a parent operates. It still follows the Kids category's rules where it can: no ads or analytics SDKs, and a parental gate before purchases, links and settings.
In short
Moonfox treats safety as part of how a story is built: rules per age band, moderation and quality checks on every story and picture with nothing unapproved shown, the child's name kept away from the models, no photos, speech kept on the iPhone, and a parent in control behind a gate. Read the full privacy policy, or ask us anything at menan@titanix.dev.