Safeguarding & safety
A large language model is not a tutor.
Point a general-purpose chatbot at a child and you don't get a tutor. You get a confident answer machine with no way to keep that child safe. When a young person's learning and wellbeing are at stake, how the AI is built is what counts.
The risks of getting it wrong
Four ways a raw chatbot can harm a child.
It does the work for them
The thing a student most often asks an AI is "just give me the answer." A tutor that gives in removes the productive struggle that builds real understanding.
It can be confidently wrong
LLMs make things up. A method that looks right but isn't, stated with total confidence, teaches the misconception instead of fixing it.
Children disclose things
Young people tell chatbots what's worrying them. Without a clear route to a trusted adult, those moments get missed. That is a serious child-protection risk, not a rare edge case.
It's children's data
Names, ages, whole conversations. Handled carelessly, that becomes a privacy failure with real legal and ethical weight.
How Saga protects
Safety built in from the start.
We answer each of those risks with a specific design decision, and we test that it actually works.
Safeguarding firewall
If a child signals distress, Saga routes it to a trusted adult through the school's safeguarding flow. Wellbeing and academics stay strictly separate, so a child's private disclosures never end up in their schoolwork record.
Privacy by construction
Students are pseudonymous. No real names, emails or dates of birth are stored or sent to the model. We follow data-minimisation and regional data-protection rules. By default the read-aloud voice runs on the device itself, so spoken lessons stay on it; a school can opt into a more natural cloud voice, which only ever synthesises the tutor's words.
No answer leakage
Questions, diagrams and worked steps are checked so they never the answer. An automated scan blocks any leak before content reaches a student.
Guardrails against error
Saga tutors from curriculum-aligned content with verified answers, so a student isn't at the mercy of whatever a model improvises in the moment.
Full transparency
Every conversation is recorded and visible to teachers. There is no black box. Staff can review exactly what was said, at any time.
Age-appropriate by default
A calm, focused interface for young people. No open-ended chat surface, and a clear, always-available way to reach a real person they trust.
When a child needs help
Academic struggle and personal distress are handled on completely separate tracks. A worried student is guided, quickly, to the right adult. Small details matter here too: the supportive message a child receives is never read out loud, so a classmate sitting nearby cannot overhear it.