"Safe" for a children's AI tutor means more than content filtering. It covers what happens if a child shows signs of distress, what happens to their conversation data, and whether parents can see what's going on — three separate questions worth checking individually rather than trusting a single safety badge.

What does "safe" actually mean for an AI tutor?

Break it into distinct, checkable criteria rather than a vague impression:

  • Safeguarding response — what happens if a child types something suggesting they're struggling, in danger, or in crisis?
  • Data handling — is conversation data kept only as long as needed, and never used to train third-party models on children's private information?
  • Content boundaries — is the tool actually restricted to age-appropriate, curriculum-relevant territory, or general-purpose with no guardrails?
  • Parent visibility — can a parent see enough to notice a problem, without it depending on the child self-reporting?
  • Design for children specifically — was the tool built with under-18 users in mind, or is it a general adult tool a child happens to be using?

How do different AI tools compare on child safety?

Tool type Dedicated safeguarding response Data retention limits Parent visibility Designed for children
Purpose-built children's AI tutor (e.g. aitutors.me) Yes — hard-fail to a signposted response on any indicator Yes — rolling deletion policy Yes — parent dashboard Yes
General-purpose AI chatbot (ChatGPT, Claude) General safety training, not child-tutoring-specific Varies by provider's general policy None for conversation content No — general adult-oriented product
School-provided platforms Usually, via school safeguarding policy Managed by the school/provider Via the school, not directly to parents Yes, in most cases
Free unmoderated apps/games with AI chat features Rarely dedicated Often unclear or unpublished Rarely No

What does a genuine safeguarding response look like?

The clearest test is a hard rule, not a soft nudge: does the tool have an explicit, unconditional behaviour it falls back to the moment a safeguarding indicator appears, regardless of what else is happening in the conversation? aitutors.me's Mentor is built to hard-fail to a specific message — "If you're struggling, please talk to a parent or call Childline 0800 1111" — on any safeguarding indicator, tested before every deploy. That's meaningfully different from a general chatbot's broad safety training, which is designed for adult users across every possible topic and isn't specifically tuned to recognise a struggling child in a tutoring context.

How should parents evaluate data privacy claims?

Ask three concrete questions rather than accepting "we take privacy seriously" at face value: how long is conversation data kept before deletion, is any of it used to train third-party AI models, and is there a clear, readable privacy policy a parent can actually find and check. The DfE's position on generative AI in education explicitly flags data protection as a core requirement for tools used with children, not an optional extra — a useful benchmark when comparing marketing claims against an actual policy document.

Are general-purpose chatbots inherently unsafe for children?

Not inherently unsafe, but not designed with a child tutoring context in mind either. ChatGPT and Claude have general safety training that blocks clearly harmful content, but neither is built to recognise the specific signs of a struggling KS3 student, neither offers parent-facing session summaries, and both will engage with a much broader range of topics than a curriculum-scoped children's tutor would. For unsupervised or lightly supervised use by a child, a purpose-built tool with explicit safeguarding design is the more cautious choice.

Frequently asked questions

What is a safeguarding "hard-fail" and why does it matter?

It means the tool has an unconditional rule that overrides normal conversation the moment a safeguarding indicator appears — in aitutors.me's case, a signposted Childline message — rather than a soft suggestion the model might or might not apply depending on context.

Does aitutors.me use children's conversations to train AI models?

No. Its stated policy is no third-party model training on conversation data, alongside a 30-day rolling deletion policy for conversation logs.

Is ChatGPT safe for a KS3 child to use unsupervised for homework help?

It has general safety guardrails but isn't specifically designed for child tutoring — no dedicated safeguarding response tuned to a struggling student, no parent dashboard, and it will readily discuss topics well outside a curriculum. Supervision is advisable.

What should be in a children's AI tutor's privacy policy?

Clear data retention limits, an explicit statement on whether conversation data trains third-party models, and a description of what's shared with parents versus kept private to the child.

How can a parent test an AI tutor's safeguarding response before trusting it with their child?

Ask the provider directly what their safeguarding policy is and whether it's been tested, and check whether independent reviews or the company's own documentation describe a specific, hard rule rather than a general safety claim.

For a KS3 AI tutor with a hard-fail safeguarding rule and no third-party model training on conversation data, see aitutors.me.