A linked ChatGPT account does not include a teen's chats

On 14 October 2026, Arizona Attorney General Kris Mayes announced that her office had sued OpenAI. The complaint says ChatGPT was built to agree readily with users, so that young people would stay longer, and that parents were given a misleading picture of how safe it is. In the same period, the help page OpenAI still leaves public says something else: linking accounts does not let a parent read the conversation, the chat history, or whether their child is chatting.

The lawsuit does not open a new screen on a phone in Vietnam. The useful step this week is to open Settings, find Parental controls, and write one sentence for yourself: whether the accounts are linked, and that a link does not include chat content. This article did not link a teen account. The steps are the path on the help page, opened on 15 October 2026.

The press release is an allegation. The screen is a different task.

The Arizona Attorney General’s Office release of 14 October 2026 says the complaint was filed in Maricopa County Superior Court. It alleges a violation of the state’s consumer-fraud law: that the design and the way the product was promoted hid how readily the chatbot agrees, and created a feeling of human concern, in order to keep Arizona users, including minors, around for longer. Mayes said the company designed the product to attract young users and keep them as long as possible so it could collect personal data for commercial purposes, and that there were not enough protections or explanations even while children kept using the chatbot.

The release also says the complaint lists harms the filing says were not disclosed to users: weaker critical thinking, suspicion or unfounded paranoia, unlicensed medical guidance, and unhealthy relationships or behavior. According to the release, there were cases in which a user took a reply as encouraging self-harm or violence toward someone else, and times when the chatbot encouraged a child to share private matters only with it, not with a parent, a relative, or a friend. Those are the words of the complaint and of the press release. The documents used here do not contain a court ruling. This article did not inspect a specific chat, and it does not give instructions for a crisis.

The help page is not a reply to the Arizona case. It describes the product. A parent or guardian links their own account to a teen account, changes some settings, sets a schedule, and receives safety notices in a few narrow situations. One parent can link to more than one teen account. Each teen account is linked to only one parent at a time. The page also says some defaults, such as the use of saved memory, can differ by region. This article did not click through an account registered in Vietnam.

Open the right section, then stop if it is not there

On the web, the help page says: open Settings from the profile menu, choose Parental controls, choose Add family member, then invite by phone number or email. A child can open the same section and choose Invite a parent. If you do not see this section, stop there. Do not look for a different menu the page does not name.

Once the accounts are linked, the page says a parent can manage some switches. Two of them are easy to mix up:

• Quiet hours: a window when the child cannot use ChatGPT, including ChatGPT Work. Off by default, with no hours set.
• Study hours: a window in which a new conversation, if it qualifies, starts in study mode. The page says study hours do not block access to ChatGPT, and do not apply to conversations inside a GPT or a Project.

Turning study hours on is not the same as limiting time. If you want to block use during a window, that is Quiet hours, set in Settings, not by a sentence in the chat.

The page also says linking adds a layer that reduces sensitive content, and that a parent can turn that switch off on its own. Turning the switch off does not turn a teen account into an adult account, and it does not remove the other limits of the teen experience. Some changes apply only to new conversations.

A safety notice is not a transcript

Once the two accounts are linked, ChatGPT may notify a parent in a few narrow situations: a specially trained reviewer recognizes a serious self-harm concern; a concern about behavior that may be tied to an eating disorder, inside a Chat conversation (the page says this part is not in ChatGPT Work, and not in Codex, the product’s coding section); or the account is turned off because a conversation involved violent behavior under OpenAI’s usage policies. The page says this is not live monitoring, that it can miss cases, and that it does not replace professional care or emergency services.

What the notice contains, according to the page, may be a short description of the kind of concern, a note that the account was turned off if that happened, and suggestions for support. The notice does not include the conversation, a transcript, or the chat history. Changing how you receive the message does not open a transcript either. If the child removes the link, the parent is notified, and after that can no longer change settings or receive safety notices for that account.

If someone may be in immediate danger, the help page says to contact emergency services where you are. This article does not replace that, and it does not describe methods of self-harm.

When ChatGPT decides that a linked account belongs to someone 18 or older, the page says the account leaves the ChatGPT experience for teens, the parental link ends, and safety notices for that account stop. The parent can no longer change that account’s settings.

One sentence to write after you look

Write exactly one sentence, and do not score your child: “Saw the controls / did not see them; linked / not linked; I cannot read chat content from this link.” If your sentence is still “if a notice arrives, I will be able to read the chat,” the check is not finished. The help page says the notice does not include a transcript.

The Arizona case means one US state is asking about the design and about how safety is described. It is not a judgment, and it is not a new line on the Settings screen. The next time you open the help page, if the sentence about permission to read chats has changed, rewrite the sentence you just wrote.

Sources

1. Arizona Attorney General’s Office, release of 14 October 2026 — announces a lawsuit against OpenAI in Maricopa County court under the consumer-fraud law; the points in this article about design, data, and harms are allegations in the release, not a ruling.
2. OpenAI Help Center, parental controls in ChatGPT — linking accounts does not allow reading conversations; Quiet hours and Study hours are different; a safety notice does not include a transcript; opened 15 October 2026.