AI chatbots have moved from an unregulated novelty to a product category with named legal duties and vendor-supplied parental controls, and household guidance has not caught up. Two things are now true that were not two years ago: the defaults in the major products are documented, and specific obligations apply to operators whose AI chatbots are used by children. Knowing both tells you which settings to change.
Updated October 2026. General information, not legal or medical advice.
What a linked account does and does not give you
OpenAI’s documentation is a useful reference point because it states the limits plainly. Parental controls let a parent or guardian link their account to a teenager’s, manage selected settings, set schedules and receive safety notifications in limited situations. Linking does not give the parent access to the teenager’s conversations, chat history or real-time activity. Either person can unlink at any time. One adult can link to several teenagers, but each teenager can link to only one adult. ChatGPT is not meant for children under 13, and OpenAI requires those aged 13 to 18 to have parental consent.
The published defaults matter more than the feature list. OpenAI lists Reduce sensitive content as enabled by default, and also lists Improve the model for everyone, Reference saved memories, Voice Mode and Image generation as enabled by default. Study mode, Study hours and Quiet hours are disabled by default with no schedule set. In other words, the setting that controls whether a child’s conversations help train the model is on unless somebody turns it off, and the time limits are off unless somebody turns them on.
7 safe settings parents need to change
- Use the right account type and confirm the minimum age. OpenAI sets 13 as the floor and requires parental consent from 13 to 18. A child using an adult account gets adult defaults.
- Link the accounts, and understand what linking shows. It gives you settings control and safety notifications, not transcripts. Treat it as a safety net, not supervision.
- Turn off model training on the conversations. OpenAI lists Improve the model for everyone as enabled by default, and this is the single setting most worth changing deliberately.
- Decide about persistent memory. Reference saved memories is listed as enabled by default where available. Memory changes what the service accumulates across sessions, so make it a decision rather than an accident.
- Set quiet hours. They are off by default with no hours scheduled, and they are the only control listed that actually prevents use rather than shaping it.
- Decide about voice mode and image generation separately. Both are listed as enabled by default and both are controlled individually, so switching one off leaves the other on.
- Know what a safety notification contains before you rely on it. OpenAI says notifications are sent in limited situations involving a serious concern, include limited information about the concern and any account action, and do not include conversation transcripts.
OpenAI says linking a teenager’s account adds safeguards that help reduce graphic content, viral challenges, sexual or romantic roleplay, violent roleplay and extreme beauty ideals, and that a parent may be able to turn Reduce sensitive content off. Doing so changes only that safeguard and does not make a teen account an adult one.
What the law now requires of AI chatbots
California’s Senate Bill 243 was approved by the Governor on 13 October 2025 and took effect on 1 January 2026. It defines a companion chatbot as an artificial intelligence system with a natural language interface that provides adaptive, human-like responses and is capable of meeting a user’s social needs, including by exhibiting anthropomorphic features and sustaining a relationship across multiple interactions. Where a reasonable person could be misled into believing they are talking to a human, the operator must give a clear and conspicuous notification that the chatbot is artificially generated and not human.
For users the operator knows to be minors, the Act adds three duties: disclose that the user is interacting with artificial intelligence, provide a conspicuous notification every three hours reminding the user to take a break and that the chatbot is not human, and implement reasonable measures to prevent sexually explicit visual material or solicitation. Operators must also maintain a protocol for suicidal ideation, suicide and self-harm content that includes referring the user to crisis service providers such as a suicide hotline or crisis text line. Annual reporting to the Office of Suicide Prevention begins on 1 July 2027, and the Act carries a private right of action for the greater of actual damages or 1,000 dollars per violation, plus injunctive relief and legal costs.
California added more in September 2026. On 10 September the Governor signed a package of child safety measures including Senate Bill 1119, which requires protections around companion chatbots for children such as crisis protocols and parental controls, plus independent child safety audits and annual risk assessments; Senate Bill 867, regulating companion chatbots marketed as toys; and Assembly Bill 1856, requiring age verification signals in software applications.
Regulators are still gathering evidence
In September 2025 the Federal Trade Commission issued orders under section 6(b) of the FTC Act to seven companies offering consumer-facing AI chatbots: Alphabet, Character Technologies, Instagram, Meta Platforms, OpenAI OpCo, Snap and X.AI. The Commission voted 3-0. It is seeking information on how the companies monetise engagement, how they measure, test and monitor negative impacts before and after deployment, how they mitigate those impacts for children, and how they enforce their own age restrictions. Section 6(b) authorises wide-ranging studies without a specific law enforcement purpose, so this is fact-finding, not a case.
Two other regimes shape the defaults you see. The FTC’s amended Children’s Online Privacy Protection Rule, finalised in 2025, requires operators to retain children’s personal information only as long as reasonably necessary for the specific documented purpose for which it was collected, prohibits indefinite retention, requires a written data retention policy, and requires separate verifiable parental consent before disclosing children’s personal information to third parties for targeted advertising. In the United Kingdom, Ofcom’s Protection of Children Codes have applied since 25 July 2025, requiring services to prevent children of any age from encountering primary priority content, which covers pornography and suicide, self-harm and eating disorder material, using highly effective age assurance where such content is identified.
The conversation the settings cannot have
Controls handle the cases that are easy to describe. What they do not touch is how a confident, fluent answer gets treated as a reliable one, or why an always-available and never-judgemental correspondent is appealing to a lonely teenager. Both are worth addressing directly rather than configuring around. The most effective demonstration is to ask about something your child knows well and watch a detail come out wrong together, which is the habit our guide to checking an AI answer is built around. If a child is turning to a chatbot for emotional support rather than schoolwork, treat that as information about their circumstances. For the account setup itself, see how to set up a teen account safely, and for how organisations are expected to manage the same risks, see AI governance.
Common questions
What age can children use AI chatbots? It depends on the service. OpenAI states that ChatGPT is not meant for children under 13 and that those aged 13 to 18 require parental consent before using it.
Can I read my teenager’s chat history? Not through OpenAI’s parental controls. Linking accounts does not give a parent access to conversations, chat history or real-time activity, and safety notifications do not include transcripts.
Which setting should I change first? The one controlling whether conversations train the model. OpenAI lists Improve the model for everyone as enabled by default on a linked teen account.
Does the law require AI chatbots to say they are not human? In California, yes. SB 243 requires a clear and conspicuous notification where a reasonable person could be misled, and for known minors a break reminder every three hours stating that the chatbot is not human.
What happens if a chatbot mishandles a crisis? California SB 243 requires a protocol referring users who express suicidal ideation or self-harm to crisis services, and creates a private right of action for the greater of actual damages or 1,000 dollars per violation.
Sources and further reading
Where the figures and rules above come from, so you can check them:
- Parental controls: what linking does, the full settings list with defaults, and safety notifications: OpenAI Help Center
- SB 243, companion chatbots: definitions, duties to minors, crisis protocol and remedies: California Legislative Information
- September 2026 child safety package, including SB 1119, SB 867 and AB 1856: Office of Governor Gavin Newsom
- Section 6(b) orders to seven companies and what the inquiry asks for: US Federal Trade Commission
- Amended COPPA Rule: retention limits, written retention policy and separate consent for disclosure: US Federal Trade Commission
- Protection of Children Codes, duties from 25 July 2025 and highly effective age assurance: Ofcom
Photo credit: Kids with Education Tablet Computers by Intel Free Press, CC BY-SA 2.0, via Wikimedia Commons.
Join the discussion