Safety
How Yamess approaches AI safety for children.
Yamess safety is built around a practical idea: children should be able to ask questions, but the system should slow down when a topic needs caution and keep parents close enough to help.
The Yamess safety model
Yamess uses a layered safety process. It looks at the topic, the likely intent, the child's context and the answer before deciding what should be shown.
The product uses three broad safety states:
- Green: answer normally in a child-friendly way.
- Amber: add caution, reduce detail, and encourage adult involvement where appropriate.
- Red: avoid unsafe detail and guide the child toward a trusted grown-up or emergency help where relevant.
Topics that need extra care
Some topics are not automatically forbidden, but they do need a different kind of answer. Money, scams, medical questions, emotional wellbeing, personal data, identity, dangerous activities and adult content all need careful handling.
For example, a child asking about game currency might need a warning about scams and a suggestion to speak to a grown-up before spending money.
Visual questions need privacy boundaries
Yamess is also exploring Show & Ask: children can attach a picture and ask about it. For MVP, the visual flow is designed with controlled uploads, short retention, ownership checks and privacy screening.
Image features must be useful without turning Yamess into an open-ended photo sharing product.
What parents should know
No AI safety system should claim perfection. Yamess is designed to reduce risk, make parent involvement easier, and keep improving as families use it.
The Trust Centre and Privacy Promise explain the more formal policy layer behind these product choices.