How families can use an AI assistant safely

Where AI helps busy families

AI is at its best as a drafter and a summarizer, not a decider. Around the house, that looks like: turning a long school newsletter into a list of dates, drafting a week of dinners from recipes you already use, writing the first version of a to-do list for a move or a birthday, summarizing a 40-message thread about the field trip, and answering "what's on Thursday?" without anyone opening three apps. It's also good at the patient stuff: explaining fractions one more time, or inventing a bedtime story with your kid as the hero.

What it's not good at is being right every time. An assistant can misread a date, invent a detail, or confidently answer a question it should have asked you. That's not a reason to avoid it; it's the reason for the rules below.

What not to share with any AI tool

The AAP's guidance on AI chatbots and kids makes a point that applies to adults too: chatbots aren't confidential, so personal information belongs only with trusted people. A household rule, for everyone:

• Never passwords, PINs, or security answers. No assistant needs them, and a good one refuses them.
• Never card numbers, bank details or government ID numbers. If a task needs a payment or a sign-in, a human does that step.
• Be careful with medical, financial and legal details. Ask general questions; keep the specifics with your doctor, bank or lawyer.
• Think before sharing other people's information. Your kid's friend's address, a teacher's phone number, a neighbor's situation: not yours to paste in.
• Assume a chat may be stored. Read the tool's privacy page. If it doesn't have a plain one, that's an answer.

Keep a human in the loop: approvals

The single most useful design question for any AI tool is: what can it do without asking me? An assistant that can read your calendar is helpful. One that can change your calendar, send an email in your name or submit a form on a website without a confirmation is a liability, however good it is, because the one time it's wrong, it's wrong in the real world.

So prefer tools where the AI proposes and you approve, especially for anything that's hard to undo: sending messages, changing shared plans, spending money, submitting anything. Small reversible actions (adding milk to a list) can be direct if there's an undo. For a family, this also keeps both parents in the loop: a proposed change can be seen and discussed; a silent one can't.

Kids and AI chatbots by age

Two sources worth reading in full. The American Academy of Pediatrics says chatbots may tell kids false, harmful, highly sexual or violent things, that companion chatbots can draw children into one-sided emotional bonds, that they don't fact-check or know the child's situation, and that younger children and kids with developmental delays, trauma or mental health concerns are more easily influenced. It doesn't set age cutoffs; it advises closer monitoring especially at younger ages, reviewing chats together, keeping AI use in shared spaces, explaining that real people offer loyalty and honesty a bot can't, and talking to your child's doctor if a kid withdraws into chatbot use. Its overall view is that children can use AI technology safely and effectively with parents involved.

Common Sense Media's age-and-stage guidance is more specific: no AI chatbots for ages 5 and under, because young children can't fully tell AI apart from real people; limited use alongside a parent or caregiver for ages 6 to 12; supervised use for homework and creativity from 13, but not for comfort or feelings; and, for everyone under 18, no AI for emotional support, mental health advice or friendship, since chatbots have no medical training and miss warning signs of conditions like depression and anxiety. Its best question for a kid is "Is this AI helping me learn, or just helping me finish?"

If you're worried about your child's mood, friendships or screen use, those sources and your pediatrician are the right place to go, not this guide.

House rules for AI

Short enough to put on the fridge, and the same for grown-ups and kids where they overlap.

• AI is for drafts, hints and ideas. People make the decisions.
• No secrets in, no secrets out: nothing private goes into a chatbot, and a parent can see any chat a kid has.
• Kids use AI in shared spaces, with a parent nearby for under-13s, and not at all for the youngest.
• Homework: ask it to explain, not to answer. "Is this helping me learn or helping me finish?"
• Feelings go to people. An AI is not a friend, a counselor or a doctor.
• Anything that changes the real world (a message, a booking, a purchase, a calendar change) gets a human's okay first.
• If it says something strange, false or upsetting, tell a parent. We'll look at it together.

How HouseChief is designed

HouseChief's AI Chief of Staff is one example of the approval model above. It can read the family's calendar, meals, recipes and lists and propose changes, but calendar and meal edits, list edits, emails, new projects and anything it would submit on a website wait for a parent's Approve; only adding an item to a list is direct, with Undo. When it runs an errand in its cloud browser, it stops before any final submit, never types into password or payment fields, and a parent can watch or take over.

For kids, HouseChief follows the age guidance by design rather than by trust. Kids have no accounts and never use the Chief of Staff. Each kid can have a Buddy instead (a parent can switch it off per kid in Settings): it gives homework hints rather than answers, tells bedtime stories and jokes, and can check that kid's chores, the calendar and the weather. It can't change anything, browse or search the web, or see parents-only data, and parents can read every chat. Whether a particular child should use it, and from what age, is still a parent's call: you can switch a kid's Buddy off in Settings → Family, and the sources above are the best guide. Everything stays in the family's household, with no ads, no trackers and data never sold; the privacy summary and the privacy and safety page have the details. If you'd rather use your own assistant, you can connect your own Claude or ChatGPT to HouseChief as a read-mostly connector that can only look things up and add to lists or the Inbox. Nothing here makes AI completely safe; the design just keeps a parent between the AI and the family's real life.