A tool that runs in your browser
AI Child-Risk Classifier
Tutor or friend? A small difference in the name, a large one in the risk. Estimated time: about 5 minutes. Your answers never leave your browser and are not stored. Evidence grade: C, an independent Dawani framework.
Start
The questions
The tool is loading.
How does it score, and what does it not claim?
The classifier asks ten questions about your design: the age band, the AI’s role, where it is used, the system’s eight capabilities, and where its answers come from. Then how its Arabic is produced, which of seven safeguards exist today, who reviews conversations, how age is known, and whether it asks for money. It weighs four dimensions: role and age, capabilities, the safeguard gap, and context and oversight. It returns one of four tiers along with what raises the risk, the missing safeguards written as design moves, and what to test with children before launch. One rule overrides the arithmetic: a companion with memory and a persona for a young band stays in the top tier however many safeguards exist. This is a design classifier, not a safety certification.
What this tool does not claim
This is a design classifier, not a safety certification or an accreditation from any body. It does not examine the system’s outputs or its training data; it reads what your team declares about its design, so a generous declaration produces a generous result. This classification does not replace testing with real children from the same age band.
Read more in tutor or friend?, or browse the rest of the tools.
The reading behind this tool
Its declared sources
These are general sources we read; they are not a certification of this tool or an endorsement of it. Responsibility for the model and its interpretation stays with Dawani.
UNICEF: policy guidance on AI for children
From it we take the direction toward telling the child what the system is, and protecting their data, their safety and being transparent with them.
Before you classify your design
Questions about the risk classifier
Are my answers about the AI’s design stored?
No. The classifier runs inside your browser, and your answers never reach Dawani or any server.
Is this an approved safety evaluation of the model itself?
No. The classifier reads what your team declares about its design. It is not an examination of the model’s outputs and not a safety certification from any body.
What is the next step after a high tier appears?
Start with the first safeguard in the "what is missing" list, and test the design with children from the same age band before any launch.
Have an AI that needs a review?
Tell us about its role in your product, and we will review with you the safeguards it needs before launch.
Updated: