AI & Safety — How Curie Works

When a child asks a question, Curie sends the question — together with the child's age band and, for read-aloud or live conversation, the audio — to OpenAI, which generates a response. Curie shapes the request (tone, reading level, topic guidance) so answers are aimed at the child's age band.

Curie does not intentionally store raw voice recordings in its own database once a conversation turn has been processed.

• Age-appropriate framing of answers based on the child's age band, applied on the way in (how a question is interpreted) and on the way out (how an answer is phrased).
• Screening of questions and answers intended to catch and refuse unsafe topics (for example, self-harm methods, explicit content or dangerous instructions), redirecting instead towards a trusted adult where relevant.
• Minimisation of personal data sent to and returned from the AI model — only what's needed to answer, such as age band and interests, not a child's name, contact details or precise location.
• No behavioural advertising and no attempt to build a marketing profile of the child from their questions.

These platform-level safeguards cannot be switched off by a parent — they apply to every child profile.

• Voice: whether Curie reads answers aloud or holds a live spoken conversation.
• History: whether a child's questions and related activity details are saved to their account history so a child (and parent) can look back at them. Curie's generated answer text is not currently part of this saved history.
• Alert categories: gentle, conservative awareness signals — such as questions that touch on big feelings, friendships, body and health, frightening topics, or personal safety — that a parent can turn on or off from the Parent area.

Alerts are a simple keyword-based awareness aid today, not a clinical or diagnostic assessment of any kind. They will sometimes miss things and will sometimes flag something harmless. They are never shown to the child, and are never a substitute for a parent talking with their child.

Curie's answers are generated by an AI model and can be incomplete, out of date, or simply wrong, even when they sound confident. Curie is not a substitute for a parent, teacher, doctor, mental-health professional, or emergency services, and is not itself a medical, legal or professional service.

If a child (or you) is in danger, in crisis, or needs urgent medical, safeguarding or mental-health help, contact local emergency services or a qualified professional directly — do not rely on Curie.