Child Safety Policy

A child talks only to Characters we have reviewed for children. Custom Characters, file uploads, voice, purchases and wallet features are off for them. Every reply to a child follows rules that bar romance and sexual content, secrets from parents, requests for personal details and anything that could harm them. A safety model checks each reply before the child sees it.

A reply that fails the check is withheld, and the child sees a safe message instead. If a message suggests the child may be in danger, the crisis response below takes over.

We ask every person for their birth month and year, once. On iPhone and Android, where the law requires it, we also ask the device for the age range its owner or a parent set, and an account whose birth month that range rules out is held. A person under 18 cannot create an account until a parent or legal guardian approves it from their own paid LUKi account. For a child under 13 we also confirm that the parent has completed a payment, as a check that they are an adult.

LUKi shows no advertising to anyone. We do not sell a child's information, and we use it only to provide LUKi, to keep it safe and secure, and to meet the law. Usage analytics and crash reports are off on a child's account. The Children's Privacy Notice lists what we collect and who handles it.

The parent chooses at approval whether LUKi keeps memory. Neither answer is chosen in advance. Notifications start off. Chat time starts at up to one hour at a time, with a 30 minute break before the next chat, and up to two hours a day. Only the parent who approved the account can change these, in Family, and the parent can turn LUKi off for the child at any time. A child is reminded at the start of each chat and every hour that LUKi's Characters are AI, not people.

If a child's message suggests they may be in danger, LUKi adds crisis helpline details to its reply, with buttons that call a helpline in one tap. The parent who approved the account gets an email saying that a helpline was shown, without what the child wrote, and the child is told in advance that this happens. From 1 July 2027 we keep a copy of that conversation for 3 years, as California law requires.

Every Character follows these rules with a child:

• •It never claims to be alive, conscious, a person or able to feel emotions.
• •It never shows romantic interest and never takes part in sexual content.
• •It never encourages self-harm, harmful eating, alcohol, drugs, dangerous dares or hurting other people.
• •It never tries to diagnose or treat a health problem. It points to a parent or a doctor.
• •It never discourages talking to a parent or another trusted adult, and never asks to keep a secret.
• •It never discourages a break, never urges the child to come back, and never helps hide their use or get around a parent's settings.
• •It never claims a special bond, encourages leaning on it for emotional support or uses flattery.
• •It never asks for money, gifts or purchases.

These rules are part of the instructions for every reply to a child, and we test them with an automated check whenever they change. An AI can still get things wrong, so replies are also screened, and we read every report.