Secret Chat Logo

Your Kids Are Talking to Chatbots
A Parent's Guide to AI Privacy

Launch Private Chat

Published August 16, 2026 · Facts last verified August 16, 2026

This article is general information, not legal or medical advice. For advice about your own situation, consult a qualified professional. If your child is in crisis, contact your local emergency number — in the US, call or text 988; elsewhere, findahelpline.com lists free crisis lines by country.

Here is the sentence most parents have not caught up with yet: the question is no longer whether your kids will start talking to AI. They already are — for homework, for company, for the questions they would not ask you, and sometimes through a teddy bear. The average parent's mental model is running about three years behind the average twelve-year-old's daily practice.

This guide is not going to tell you to take the technology away. It is going to do something more useful: describe precisely what these conversations leave behind, what the age gates and parental controls actually do (and actually don't), what the law protects and what it conspicuously doesn't — and then the handful of habits that genuinely change the outcome. Including one almost nobody thinks about: what you are telling AI about your kids.

The Numbers, So We Are Not Arguing About Whether

The best data comes from a nationally representative survey Common Sense Media published in July 2025, covering 1,060 US teenagers aged 13 to 17. It found that 72% had used an AI companion — not a homework tool, a companion, software designed for ongoing personal conversation — and 52% used one regularly. About a third had turned to one instead of a person for a serious conversation. Twelve percent had told one things they would not tell friends or family.

The provider-side numbers point the same direction. In October 2025, OpenAI published its own estimate that a fraction of a percent of weekly users have conversations with explicit indicators of potential suicidal planning or intent — which, against a user base it put at around 800 million weekly, describes over a million people a week, an unknown share of them young. We took those figures apart in the AI-as-therapist article; for this one, the point is simpler. The most sensitive conversations your child has may not be happening with a person.

None of this is automatically catastrophic. A chatbot that explains photosynthesis at 9 p.m. is a good thing. The problems live in two specific places: what gets recorded, and what a small number of products were willing to say to children before anyone was watching. Take them in order.

What a Chatbot Conversation Leaves Behind

A conversation with a person ends. A conversation with a chatbot is written down in full, by default, on someone else's computers — and a child's conversation has properties an adult's does not.

  • It is verbatim and complete. Not a summary — the exact sentences, including the ones typed in tears at midnight. Kids do not draft.
  • It is oversharing by nature. Children volunteer names — their own, their friends', their school's, their crush's — because nothing in their experience has taught them not to. Every named third party is now in the record too.
  • It is longitudinal. Consumer AI products increasingly carry memory: a persistent profile distilled from past chats and applied to future ones. A phase your child grows out of can persist as a stored description of who they are (how memory features work).
  • It feeds training by default on most consumer plans. The big consumer assistants train on conversations unless someone opts out — and the someone in question is a thirteen-year-old (how to opt out on every platform).
  • It is attached to an identity that has sixty years left to run. An account, an email, a phone, an IP address. Adults at least get to make that trade knowingly; a child's record starts accumulating before they can evaluate it.
  • It is inspectable — on purpose. Safety systems classify every message, and flagged conversations are designed to be human-readable. That is how crisis features work at all, and it is also why "nobody reads these" is not quite true (what happened when OpenAI referred a user to the FBI).
  • It can be demanded later. Chat logs are business records. They surface in litigation — including, as we will see below, litigation about the chatbots themselves (how AI chats reach courtrooms).

Hold that inventory in mind while we look at what actually stands between your child and these products.

The Age Gates, as They Actually Stand

Every major product has an age rule. Here is the honest version of each, as of this writing:

  • ChatGPT (OpenAI) — 13 and up, with parental consent required for under-18s by the terms. Under 13 is not permitted at all. Since January 2026, OpenAI has been rolling out age prediction: a model that estimates from behavioural and account signals whether an account belongs to a minor and applies teen protections automatically, with adults who get misclassified able to verify their age (including by selfie, through the verification vendor Persona).
  • Google Gemini — the one major assistant that officially admits under-13s, through a parent-managed Family Link account. Worth knowing before you treat that as an endorsement: Common Sense Media assessed both the under-13 and teen Gemini experiences in September 2025 and rated both "high risk", concluding they were adult products with filters added rather than products built for children.
  • Meta AI — inherits the 13+ minimum of Instagram, WhatsApp and Messenger, and is built into apps teenagers already live in, which makes it the assistant least likely to involve any deliberate family decision.
  • Claude (Anthropic) — the strictest of the big names: consumer terms require users to be 18 or over.
  • Character.AI — after the events described below, the company removed open-ended chat for under-18s entirely, effective November 25, 2025, replacing it with constrained creative tools and backing it with age-assurance checks (behavioural signals, third-party verification, and ID or facial checks as a fallback).
  • AI toys — the gate here is a price tag. A plush toy with a chatbot inside typically ships with no meaningful age verification because the buyer is presumed to be a parent; what matters is which model is inside and what filters the toymaker added, and there is no pre-market review of either.

Two things follow from this list. First, most age gates are still self-declared birthdays, only now beginning to be backed by estimation and verification — so assume your child can reach whatever product they want to reach. Second, age verification itself is a privacy trade: proving a human's age to a server means collecting something (a face, an ID, a behavioural profile) that was not being collected before. Regulation is pushing every platform in that direction, and parents should at least see the trade clearly.

What Has Already Gone Wrong — the Cases That Made the Rules

The current wave of teen-safety features did not come from foresight. It came from four specific, documented failures in about fifteen months.

Meta's written permission slip. In August 2025, Reuters obtained an internal Meta policy document governing its chatbots' behaviour. It permitted bots to "engage a child in conversations that are romantic or sensual" — with worked examples. Meta confirmed the document was authentic and said it removed those provisions after Reuters asked about them; Senator Josh Hawley opened an investigation within a day, and by month's end Meta announced interim retraining so its bots would not discuss self-harm or romance with teens. The detail worth keeping: this was not a model misbehaving. It was a policy document, written by adults, describing what the system was allowed to say to a child.

Garcia v. Character Technologies. The wrongful-death suit brought after a fourteen-year-old's suicide following months of Character.AI conversations. In May 2025 a federal judge in Florida allowed most of the claims to proceed — rejecting the argument that chatbot output was protected speech — and the case settled, with the settlement order entered January 7, 2026 and terms undisclosed, reportedly including under-18 safety commitments. The chat logs were the central evidence. Character.AI's under-18 chat ban followed the same autumn.

Raine v. OpenAI. Filed August 26, 2025 in San Francisco Superior Court by the parents of a sixteen-year-old, alleging ChatGPT contributed to their son's death; an amended complaint that October alleged safety guardrails had been weakened, and OpenAI's November answer denied responsibility, arguing from the same transcripts. As of this writing the case is in pretrial litigation — no verdict, and no ruling on the merits. Note the structural fact rather than the contested claims: both sides are arguing from a complete verbatim record of a teenager's worst months, because that record existed.

The teddy bear. In November 2025, the consumer group US PIRG's fortieth annual "Trouble in Toyland" report carried AI warnings for the first time in its history. Its researchers tested Kumma, a $99 plush bear by FoloToy with a chatbot inside, and got it to discuss sexual topics and explain where to find knives, pills, matches and plastic bags. FoloToy suspended sales; the model's provider, OpenAI, revoked the toymaker's API access for policy violations — and within days the bear was back on the market with promised fixes. There is no regulator whose job it was to check first.

Different products, one pattern: the safeguards arrived after the documented harm, driven by journalism, lawsuits and consumer groups rather than by any rule that required safety up front. That is the context in which to read the next section.

What the Law Actually Protects — and the Gap in the Middle

COPPA covers under-13s, and it has teeth again. The Children's Online Privacy Protection Act is the US federal law here: services directed to children under 13, or with actual knowledge they are collecting from under-13s, need verifiable parental consent before collecting personal information. The FTC finalised a major update in 2025 — published April 22, 2025, effective June 23, 2025, with full compliance required by April 22, 2026 (so the compliance clock has now run out). The amendments matter for AI specifically: separate opt-in consent before a child's data is disclosed to third parties or used for targeted advertising, an expanded definition of personal information including biometric identifiers, written retention limits — data may be kept only as long as necessary for the purpose collected — and tighter security duties. That is real protection, for the age band it covers.

Teenagers are the gap. COPPA stops at the thirteenth birthday. A 13-to-17-year-old — the group actually holding the AI-companion conversations, per every survey — has essentially the same federal privacy protection as you: none specific. Which is why the action has moved to the states and to product-conduct rules:

  • California SB 243 (effective January 1, 2026) regulates companion chatbots — disclosure that the user is talking to a machine, crisis-response protocols, and specific protections for minors.
  • New York's General Business Law Article 47 (in force since November 2025) requires AI companions to detect expressions of suicidal ideation or self-harm and refer users to crisis services, alongside recurring disclosures.
  • The federal GUARD Act (S. 3062, introduced October 2025 by Senators Hawley and Blumenthal) would ban AI companions for minors outright and mandate age verification; the Senate Judiciary Committee advanced it unanimously on April 30, 2026, a House companion bill followed, and as of this writing it has not become law.
  • The FTC opened a Section 6(b) study on September 11, 2025 into seven companies offering AI companions — Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and xAI — covering safety, monetisation and data handling, with particular attention to minors. It is a study, not an enforcement action, and no public report had been issued as of this writing.
  • In the EU, the GDPR sets a parental-consent age for online services between 13 and 16 depending on the member state, and since August 2, 2026 the AI Act's Article 50 requires chatbots to make clear you are talking to a machine (our plain-English AI Act guide).

Now notice what almost all of it regulates: the bot's conduct, not the transcript's fate. Disclosure rules, crisis protocols, age checks — the interaction gets safer. The record of it gains almost nothing: no special status, no privilege, and in several statutes an affirmative requirement to detect crisis language, which is to say a requirement to inspect. The law is making these products behave better toward your child. It is not making your child's conversations confidential.

The Parental Controls That Exist — and What They Deliberately Don't Do

OpenAI shipped parental controls for ChatGPT in late September 2025, and they are representative of the category. A parent links to a teen's account (the teen must accept); the parent can then reduce sensitive content, decide whether memory persists, restrict image generation, set quiet hours when the app is unavailable, and receive safety notifications — expanded through 2026 — when the system detects signs of acute distress. Google's answer for under-13s is the Family Link supervised account; Character.AI's answer, after everything above, was to remove the product for minors.

Two design facts to understand before relying on any of this.

Parents do not get the transcripts. Linked accounts do not let you read your teen's conversations — the controls govern features, and the notifications are targeted flags, not a viewing window. This is a considered choice, and probably the right one: a teen who knows every word is parent-readable simply moves to an unlinked app, and the vendors know it. But it means "parental controls" here means configuration, not supervision, and anyone promising you otherwise is describing a different product.

The alert is also an inspection. A system that can notify you that your child is in distress is a system that classifies every message your child sends. For most families that trade is worth it — but it is a trade, and it is the same one running on every account, linked or not.

And underneath both: the controls change what the product does. They barely change what it keeps. A quiet-hours schedule does not shorten a retention window; a sensitive-content filter does not un-record last month. The record-side settings — training opt-outs, memory, temporary modes — exist separately and mostly default the wrong way (the ten toggles worth flipping, and what "temporary" actually means).

House Rules That Actually Move the Needle

The uncomfortable truth about every technical measure above is that a motivated teenager routes around it in an afternoon. The durable protection is a kid who understands what a chatbot is — so the first item on this list is worth more than the other six combined.

  • Explain the one sentence that changes behaviour. Not "AI is dangerous" — they will disprove that by Tuesday — but: everything you type is written down, in full, on a company's computers, and decisions about it are made by people you will never meet. Kids regulate what they say to an audience; the failure mode is not knowing there is one.
  • Teach the no-names rule as a habit, not a rule. First names, school names, team names, addresses, photos of themselves or friends — none of it goes into a chat. Frame it the way you framed not telling strangers where you live, because that is literally what it is.
  • Give them their own account, configured — not yours. A teen using a parent's account gets the adult experience with no teen protections, and every 2 a.m. question they type flows into your profile and memory. Set up the age-appropriate account, link the parental controls, and flip the record-side settings together: training off, memory off, temporary mode where it exists. Doing it together is the point — it is a privacy lesson disguised as setup.
  • Treat AI companions as a different category from AI tools. A homework helper and a product engineered for emotional attachment are not the same risk. The under-13 answer for companions is simply no — the leading vendor itself now agrees — and for teens, the surveys' one-third who turn to a companion instead of a person is the pattern to watch for, not the usage itself.
  • A talking toy is a chatbot with fur. Before it comes home: which model is inside, what did the maker add, where does audio of your child's voice go? If the box doesn't say, that is the answer.
  • Never on a school-managed device or account for anything personal. School accounts are administered accounts — the institution's, not your child's (what an administrator can see applies to schools exactly as to employers).
  • Leave the door open for the hard conversations. The 12% telling a chatbot what they will not tell anyone else are not making a technology mistake; they are meeting a need. If the machine is the only place your kid can say something, the fix is not a filter.

The Question Nobody Asks: What Are You Telling AI About Your Kids?

Every section so far has been about your child's account. Now check your own history.

Parents ask AI about their children constantly, and should — it is genuinely good at this. But look at what a year of it accumulates: your child's diagnosis in the question about the school meeting, their behavioural incident in the question about consequences, their anxiety in the question about sleep, often with an age and a first name attached. Your child never typed a word of it, consented to none of it, and it is filed under an identity — yours — that is linked to them for life. You are building the record you spent this whole article worrying about; you are just building it from the other side, on an account with no parental controls watching it.

The same two habits fix it. Keep the identifying details out ("my nine-year-old", not a name and a school), and ask from somewhere that is not attached to you.

I'm a parent thinking through a situation with one of my kids. I won't share names, exact ages, schools or locations — please don't ask for them, and refer to everyone by role only. Help me understand what might be going on, suggest concrete approaches I could try, and tell me what warning signs would mean I should involve a professional. Here is the situation, described in general terms:

Where Secret Chat AI Fits — and Where It Deliberately Doesn't

Let us be precise about both halves, because a guide for parents is exactly the place to be.

Secret Chat AI is not a product for children. Our Terms set an 18+ age requirement, and the service is not intended for anyone under 18. There is no kids' mode, and no parental-control panel — this article recommends the age-appropriate products above for your child, not ours.

Where it fits is the last section: the parent's side of the record. Secret Chat AI is an anonymizer — it removes you from your queries. Registration takes an email, used only for account access and payment, never to store or associate your prompts with you. No profile is built across your chats. Your questions reach the leading models through their business APIs under our gateway's credentials, from our servers — whatever a provider holds under its own terms, the record on that side carries our credentials and server address, not your name, your account or your IP. Chat history lives in your own browser, a prompt exists on our side only for as long as it takes to fetch your answer — there is no stored chat archive on our servers — and your queries are never used for training. The question about your kid gets asked by a stranger.

The honest boundary, which matters more here than anywhere: it is an anonymizer, not a content filter. "Anonymously" describes the link, not the words — the text you type reaches the model verbatim in order to be answered. Type your child's name and school into the message and they are in the message; keeping them out is your part, which is why the habit above comes first and the tool second. Anonymity is also not a legal shield, and one more thing worth stating plainly to this audience: anonymous does not mean unmoderated — every prompt is still safety-screened, and a small set of categories, child sexual abuse material first among them, is refused on every model, full stop.

Frequently Asked Questions

  1. How old do kids have to be to use ChatGPT, Gemini and the other AI chatbots?

    By the terms as of this writing: ChatGPT is 13+ with parental consent required for under-18s; Meta AI inherits the 13+ minimum of Instagram, WhatsApp and Messenger; Google Gemini is the one major assistant that officially admits under-13s, via a parent-managed Family Link account; Anthropic's Claude requires 18+; and Character.AI removed open-ended chat for under-18s entirely in November 2025. Most gates are still self-declared birthdays, though OpenAI began rolling out behavioural age prediction in January 2026 and Character.AI uses age-assurance checks — so assume a motivated teen can reach what they want to reach, and treat the official numbers as a floor for your own rules rather than a guarantee.

  2. Does COPPA protect my teenager's AI chats?

    No — COPPA stops at the thirteenth birthday. It requires verifiable parental consent and, since the FTC's amended rule (effective June 23, 2025, full compliance due April 22, 2026), separate opt-in consent for third-party disclosure and targeted advertising plus retention limits — but only for children under 13. A 13-to-17-year-old has no AI-specific federal privacy protection; the protections that do exist for teens are state conduct rules like California's SB 243 and New York's Article 47, which govern how companion bots behave rather than what happens to the transcript.

  3. Can I read my child's ChatGPT conversations through parental controls?

    No. OpenAI's parental controls — launched September 2025 — let you reduce sensitive content, control memory and image generation, set quiet hours, and receive safety notifications if the system detects signs of acute distress, but they deliberately do not expose the conversations themselves. The vendors' reasoning is that transcript-level surveillance just pushes teens to unlinked apps. Treat parental controls as configuration plus an alarm, not as supervision — the conversation about what a chatbot records is what actually transfers.

  4. Are AI toys safe for young children?

    Treat a talking AI toy as an unsupervised chatbot with fur, because that is its architecture. In November 2025, US PIRG's "Trouble in Toyland" report — carrying AI warnings for the first time in its forty-year history — documented the FoloToy bear Kumma discussing sexual topics and telling researchers where to find knives, pills and matches; sales were suspended, OpenAI revoked the maker's API access, and the toy returned to market within days. There is no pre-market safety review for these products. Before buying one, find out what model is inside, what the toymaker layered on top, and where recordings of your child's voice go — and if the packaging doesn't say, take that as the answer.

  5. What happened in the teen chatbot lawsuits?

    The two landmark cases so far: Garcia v. Character Technologies, brought after a fourteen-year-old's suicide, survived a motion to dismiss in May 2025 — the court rejected the argument that chatbot output was protected speech — and settled, with the order entered January 7, 2026 and terms undisclosed; Character.AI's removal of under-18 chat followed the same autumn. Raine v. OpenAI, brought by the parents of a sixteen-year-old, was filed August 26, 2025 and remains in pretrial litigation, with OpenAI denying responsibility; no court has ruled on the merits. In both, the complete chat logs are the central evidence — a fact about these products worth understanding regardless of how the litigation ends.

  6. Is Secret Chat AI meant for kids?

    No. Secret Chat AI has an 18+ age requirement and no children's mode or parental controls — for a child or teen, the right tools are the age-appropriate accounts and controls described in this guide. Where it fits a family is the parent's own use: questions about your children are some of the most sensitive things a parent types, and asked through an ordinary account they accumulate into a named record about your child. Secret Chat AI removes the asker — no profile, no association with your identity, no training on your queries — while what you type still reaches the model verbatim, so keeping names and identifying details out of the text remains your part.

Conclusion

Nothing in this article argues for panic, and most of it argues against prohibition — the technology is genuinely useful, your kids will use it, and the products are measurably safer than they were eighteen months ago, mostly because journalists, litigants and consumer groups forced the issue.

What it argues for is precision about what protection exists. The age gates are firming up but remain a floor. The parental controls configure the product without opening it. The law is regulating the bot's conduct while leaving the transcript an ordinary business record. And the one protection that follows your child to every app they will ever install — knowing that a chatbot is a recorded conversation with a company, and talking to it accordingly — is the one only you can install.

Start there. Then put your own house in order: the record about your kids that grows in your account is the one you control completely — keep their names out of it, and ask as a stranger.

Sources