Is AI tutoring safe for children? A guide for Australian parents

Why this needs a careful look

AI is already part of many children's lives. In 2026 the eSafety Commissioner surveyed Australian children aged 10 to 17 and found that most had used an AI assistant, that a meaningful minority reported a potentially inappropriate or harmful interaction, and that many had shared personal or potentially sensitive information with one. Read eSafety's Talking to machines report for the full findings.

Those figures cover a broad mix of AI assistants and companion products, not AI tutors alone. They should not be used to suggest every product carries the same risk. They do show why a tool's purpose and safeguards matter.

An education tool should help a child learn without presenting itself as a friend, counsellor or all-purpose authority. eSafety stresses that AI is not a trusted adult and cannot replace human support for wellbeing. The national Australian Framework for Generative AI in Schools also identifies errors, bias, privacy, security and inappropriate use as risks that need active management.

An eight-part safety checklist for parents

1. Is the product designed for the child's age and purpose?

Start with the provider's minimum age and account rules. A general AI service may not permit primary-aged users to hold an account. A child-specific or school product should explain how younger users gain access and how responses change with age.

PocketTutor supports Foundation to Year 12. Students in Years 9 to 12 may create their own account; Years 7 and 8 must confirm they are at least 13; younger students are directed to a parent, who creates and manages their profile. The child needs no login - only a profile inside the adult account. Every response is calibrated from the student's year level: vocabulary, sentence length, examples and depth change across age bands. That is more specific than a claim that content is simply “family friendly”.

2. Does it stay within a clear role?

A tutoring product should be able to describe its boundaries. PocketTutor is limited to school subjects and study support. It declines sexual content, gratuitous violence, hate speech and harm-enabling requests, while allowing age-appropriate treatment of legitimate curriculum topics such as war history, Shakespeare or senior biology.

It is not a doctor, counsellor or crisis service. Buddy mode can support study motivation and exam nerves, but the app should not replace a trusted adult or qualified health professional.

This difference matters because eSafety distinguishes functional AI assistants from companion products designed to simulate relationships, even though the boundaries can blur. PocketTutor has no user-to-user messages, chat with strangers or advertising, and its AI tutors are identified as AI.

3. What happens if a child appears distressed?

Look for a concrete answer, not “our AI is empathetic”. A system should prioritise safety, direct the child towards real people and avoid improvising as a mental-health professional.

PocketTutor runs a safety check on every free-text message. If a message suggests self-harm, abuse, an eating disorder, bullying or serious distress, normal tutoring stops. The app displays a short, reviewed response that encourages the child to speak to a trusted adult and points to Kids Helpline, with other Australian services such as Lifeline, 1800RESPECT, Butterfly Foundation or eSafety where relevant. These responses are written and reviewed in advance rather than generated in the moment, and the classifier behind them is tested against written cases before release. No classifier catches every possible message in the real world.

4. How much child data does it ask for?

The Office of the Australian Information Commissioner advises parents to know which services a child uses, what information they provide, and how privacy controls are set. Review the privacy policy before uploading worksheets or starting conversations, and teach children not to enter other people's personal information.

PocketTutor's student profile asks for a first name, state, year level, subjects, learning goals and preferences. It does not ask for the child's date of birth, school, address, phone number or separate login. A parent can delete one child's profile without deleting the household, or delete the whole account in Settings.

PocketTutor uses paid AI API services rather than consumer chatbot accounts, and states that the AI providers used for tutoring, diagrams and speech do not use customer data to train their models. Data is processed overseas, mostly in the United States, so PocketTutor does not claim Australian data residency. The Privacy Policy lists each processor and where it operates.

5. Can a parent set boundaries without reading private conversations?

Different families will choose different levels of oversight. Look for controls that match the child's age and the sensitivity of the activity.

PocketTutor lets parents manage state, year, subjects, tutor, goals, study times and topics they do not want a child to explore. Profile PINs separate siblings on a shared device. On Plus, parents see activity, study habits, curriculum coverage and progress - not conversation transcripts. Sessions in which a child received a distress response do not appear in the activity trail.

That privacy boundary is deliberate. It gives parents information for a constructive check-in without turning every learning question into surveillance.

6. How does the product handle wrong answers?

Generative AI can be convincing and wrong. A safe learning environment should state that visibly, make feedback easy and use extra checks where the consequence of an error is higher.

PocketTutor grounds subject-based chat and quizzes in the relevant curriculum or senior syllabus content. Multiple-choice answers are marked in code. Generated Maths answers and multiple-choice quiz quality are re-checked; a question that fails is withdrawn and replaced. Students can report any response, quiz question or image with one tap, and the app displays a standing reminder to verify important information.

These measures reduce risk; they do not guarantee correctness. Encourage your child to compare important answers with class notes, prescribed texts or their teacher.

7. Are images, uploads and free-text inputs covered?

Safeguards should apply wherever a child can enter or receive content - not only in the main chat. Ask whether checks cover document quizzes, photos, uploads, generated images and feedback surfaces.

PocketTutor applies its free-text checks across chat, quizzes, document-based quizzes and study capture. Educational images pass through a checker that fails safe, showing no image if it cannot approve one. Generated diagrams have standing age-appropriate rules and exclude real people's faces and violent content.

Students should still avoid uploading documents containing names, addresses, school identifiers, medical information or another person's work unless that sharing is appropriate and permitted.

8. Can problems be reported and addressed?

A report button is only useful if there is a review process behind it. PocketTutor allows one-tap reporting on answers, questions and images. Inappropriate-content reports alert the team for review, and failures can be added to future test sets. Families can contact us about support or data requests.

Before choosing any provider, also check whether it publishes contact details, a privacy policy, terms, a deletion path and a clear account owner.

How to introduce an AI tutor at home

Even a tool designed for independent study benefits from a short family setup.

1. Try it together first. Use a real, low-stakes homework question and notice whether the tool explains the method.
2. Set privacy rules.Do not share a full name, school, address, location, passwords, private images, health details or another person's information.
3. Set learning rules.Attempt the work, ask for help, verify important answers and follow the school's AI policy.
4. Name the human fallback. A teacher for school rules and difficult subject questions; a parent or trusted adult for personal issues; emergency and specialist services when safety is involved.
5. Check in without judgement. Ask what the child learned, what seemed wrong or uncomfortable, and whether they reported it.

Queensland's Department of Education gives similar parent guidance: balance AI with other forms of learning, protect personal information, use reputable platforms, check age terms and review outputs critically. See its generative AI guidance for families.