Child safety

Who sets up an account

• In place today. A parent or guardian creates the account. A child cannot sign up. When a parent registers, they say they are the parent or legal guardian of the child they will add.
• In place today. A parent adds each child's profile. We do not ask for a child's email address, and a child's profile is identified by a random code, not by their name.
• In place today. A parent's agreement to the terms and the privacy notice is recorded with its version and the time. A parent can withdraw it, and a profile cannot be added while it is withdrawn.
• In place today. A child uses a shared family device with a PIN that a parent sets. A parent can set study windows and can remove a device.
• Not decided yet. We do not check a parent's identity or age beyond this. Whether that is enough for the places we plan to launch is a decision for the founder and a lawyer.

The coach is an AI

• In place today. Wherever the coach talks to a child, the screen says it is an AI and not a person, in plain text that a screen reader reads aloud.
• In place today. The coach is built not to claim to be a person, ask for personal details, ask a child to keep a secret from their parent, or move the conversation off the app.
• In place today. The AI does not decide what a child knows. Progress is recorded from the child's own answers to independent checks, and it is described as an estimate.

Checks on what is said

• In place today. What a child types, and the text read from a photo, is checked by written rules and a second AI check before the coach answers. What the coach writes is checked again before the child sees it. If a check cannot finish, the turn is stopped, not let through.
• In place today. If a message suggests a child may be at risk, the coach stops the lesson and shows a short, kind message with helplines. The event is recorded. A parent can see on the child's coach page that a pause happened, and when, but not the child's words.
• Built, not switched on yet. Telling a parent by email when a pause happens is built but switched off, because no email service has been chosen yet. Until then a parent sees pauses only in the app.
• Not decided yet. The safety rules were written by us and have not been reviewed by a child-safeguarding specialist yet. That review is planned before the product is released to families.

Data and advertising

• In place today. There are no ads, we do not sell data, and we do not train AI models on children's data by default. We have not built any of these and do not plan to.
• In place today. A parent can ask for a copy of their family's data or for it to be deleted, and can choose how long chat and problem text is kept.
• Planned, not yet in place. The AI models are planned to run on Amazon Bedrock in European regions, and the app and database on a server in the EU. Until that is set up and written into our privacy notice, do not read this page as saying where your child's data is today.

Contact and reporting

• Not decided yet. A contact address for child-safety questions and concerns is not set up yet. It will be added here before any child uses the product.

What this page does not say

It does not say the coach can never make a mistake, and it does not say that any law is complied with. If something on it is out of date, the product's own behaviour and the privacy notice come first. We keep a dated record of how each of Anthropic's published requirements for products used by young people is met, and what is still open.