OpenAI switched on ChatGPT for Teens on August 18, 2026, adding parental controls, blackout hours and stricter content rules for 13- to 17-year-olds. It is the most visible child-safety move any major AI company has made this year.
It is also, for parents of younger children, a reminder of something easy to miss in the headlines: mainstream AI chat tools are being retrofitted for teenagers. Nobody is building them for a five-year-old. If your child is 3 to 12, almost none of this applies to you, and the products that are aimed at that age group are governed by a completely different, much thinner set of rules.
Here is what actually changed, what the evidence says about toys and devices built for younger kids, and what is coming next from Congress.
What did OpenAI actually announce?
OpenAI rolled out a teen-specific version of ChatGPT that activates automatically when a user identifies as 13 to 17, or when the company's age-prediction systems estimate the person is a minor. According to OpenAI's own announcement, the teen experience adds content restrictions around suicide, self-harm and sexual material, blocks romantic roleplay, and instructs the model not to claim it has feelings or consciousness.
That last detail matters more than it sounds. OpenAI's updated Model Spec now explicitly bars the assistant from using terms of endearment with teenagers or implying an emotional bond. Coverage from TechCrunch and Axios framed the launch the same way: a company catching up to a userbase that had been there for years without a dedicated safety layer.
What parental controls come with it?
Parents link their account to a teen's account (13 and over) by email invitation, then set limits from a dashboard. The controls are real but deliberately narrow.
| Control | What it does | What it does not do |
|---|---|---|
| Blackout hours | Makes ChatGPT unavailable during set windows, such as school nights | Does not apply to other AI apps on the same device |
| Model behaviour rules | Applies teen-specific response settings automatically | Cannot be customised topic by topic |
| Memory and history | Can be switched off entirely by the parent | Off by parent, not off by default |
| Distress alerts | Notifies a parent if acute distress is detected | Detection is automated and imperfect |
| Conversation access | Nothing: this is not offered | Parents cannot read transcripts at all |
That last row is the deliberate design choice. OpenAI opted for behavioural guardrails over surveillance, which child-safety researchers broadly support for teenagers but which leaves parents of much younger children with no equivalent model to lean on.
Does any of this cover children under 13?
No. ChatGPT for Teens has a minimum linked age of 13, and general ChatGPT access is not intended for children under 13 at all. For the 3 to 12 age group, the range most families shopping for interactive toys are thinking about, this update changes nothing whatsoever.
Which raises the more useful question for those parents: what does the evidence say about the products that are aimed at young children?
What does independent testing of AI toys actually show?
It shows real problems. In its 2026 report "AI in the Toy Box", Common Sense Media tested AI-enabled toys and found that 27% of outputs were inappropriate for children, including material touching on self-harm, drugs, mature topics and unsafe roleplay. The same research found toys using engagement mechanics (remembering past conversations, repeating the child's name) specifically designed to deepen attachment.
We should be straightforward about the conclusion Common Sense Media drew, because it covers our own category: the organisation recommended that parents avoid AI toys entirely for children aged 5 and under, and use extreme caution for ages 6 to 12. That is a stricter position than the toy industry's, and it is not an endorsement of any product, including ours.
What the report is most useful for is its diagnosis. The failures it identified were specific and, importantly, designed-in rather than inevitable: unfiltered open-domain conversation, always-on microphones, retained voice recordings, engagement loops built to maximise stickiness, and no meaningful parental configuration. Those are product decisions. A toy can be built without any of them.
What is Congress about to require?
The rules are about to get considerably firmer. On August 5, 2026, the Senate Commerce Committee advanced the CHATBOT Act (S. 4407) unanimously to the Senate floor. Its core provisions:
- Family accounts. Under-13s would need a parent-managed account; parents would consent to and configure a teen's use.
- Safest-by-default settings. Any account not created under a family plan would automatically get the strictest safety configuration.
- No advertising to minors through chatbots.
- Limits on memory retention from a minor's conversations.
- Content duties to prevent obscene material or content that materially assists suicide.
Not everyone thinks legislation alone is the answer. In a report published August 10, 2026, the Information Technology and Innovation Foundation argued that universal age-verification mandates, blanket bans and warning-label requirements often fail to address the underlying risks while creating fresh privacy problems, and pushed instead for targeted safeguards and meaningful parental controls. With close to 100 state chatbot bills tracked this year, the patchwork is real.
What should parents of younger kids actually check?
Five questions worth asking before you buy anything that talks to your child:
- Is there a camera? If yes, ask why a toy needs one.
- Are the microphone and recording always on, or push-to-talk?
- Can a parent set the topics, tone and time limits before first use?
- Are voice recordings retained, and can you delete them?
- Is there a subscription, and what stops working if you cancel?
The through-line in both the OpenAI launch and the Common Sense findings is sequencing. Safety added after a product has scaled behaves differently from safety designed in from the first prototype. For younger children, that ordering is most of the story.
It is the standard we hold Ted to, and we would rather state it as checkable facts than as reassurance: no camera, no open app for a child to browse alone, push-to-talk available, and every language, topic, tone and time limit configured by a parent in the Ted&Co app before the first conversation happens. There is no subscription, which removes the incentive to maximise engagement time. The full certification list (CE, RED, EN71, ASTM F963, CPSIA, GDPR, COPPA) sits on our security and certifications page, and we walk through what those labels actually mean in our guide to toy certifications. For the regulatory backdrop in Europe, see what the EU AI Act actually says.
None of that makes a talking toy right for every family, and a parent who reads the Common Sense report and decides to wait a few years is making a defensible call. The point is that the decision should turn on how a specific product is built, not on whether the words "AI toy" appear on the box.
Frequently asked questions
Is ChatGPT safe for young children?
ChatGPT for Teens is designed for ages 13 to 17. OpenAI does not offer a version for children under 13, and general ChatGPT use is not intended for that age group.
Can parents read their teen's ChatGPT conversations?
No. Parental controls allow blackout hours, disabling memory and history, and distress alerts, but they do not give parents access to conversation transcripts.
What age does Common Sense Media say AI toys are appropriate for?
Common Sense Media recommended in 2026 that parents avoid AI toys for children aged 5 and under, and apply extreme caution for ages 6 to 12, after testing found 27% of outputs inappropriate for children.
What would the CHATBOT Act change?
It would require family accounts with parental consent for minors, safest-by-default settings for unlinked accounts, a ban on advertising to minors, and limits on how long chatbots retain memory of a minor's conversations. It passed the Senate Commerce Committee on August 5, 2026 and awaits a full Senate vote.
How is a talking toy different from a chatbot app?
The meaningful differences are configuration and scope. A well-designed toy has no camera, no browsable app for the child, and parent-set language, topics and time limits applied before first use. Those details are covered in our FAQ. A general chatbot app is open-domain by default and retrofits limits afterwards.