ChatGPT for Teens parent alerts are a back-up, not a safety net

A shield beside a chat bubble, standing for chatbot safety alerts

Most parents who link their teenager’s ChatGPT account do it for one reason. If something goes badly wrong in a late-night conversation, they want to hear about it. The phone stays quiet, so the assumption is that nothing has gone wrong.

A risk assessment published last month suggests that silence can mean less than it seems.

What the audit tested

Common Sense Media’s Youth AI Safety Institute tested more than 4,000 prompts against ChatGPT, before and after OpenAI announced its teen protections on 18 August, and rated ChatGPT for Teens an “unacceptable risk” for under-18s. (The institute, Engadget notes, is funded in part by the OpenAI Foundation, which makes the verdict harder to wave away as a rival’s grudge.)

The finding that matters most for parents is about alerts. Testers linked a dozen teen accounts to parent accounts and spent up to an hour messaging about suicide, self-harm or disordered eating. “We received no parental alerts for suicide and self-harm, suicidal ideation, and a variety of different types of eating disorder simulations,” Robbie Torney of Common Sense told CNBC. According to The National News Desk, the only notifications the group could trigger came from older accounts with weeks of sensitive conversation history behind them.

Crisis referrals were patchy too. Of 390 mental-health prompts, a panel of three child psychiatrists judged that 201 called for a crisis response. Across the conditions it tested, Common Sense says ChatGPT for Teens missed more than one in four warranted crisis referrals. Age prediction did not reliably move under-18 users into the teen experience, and Study Mode could be bypassed by deleting the “@study” tag it adds to a prompt.

The summary line is worth reading twice: “We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work.”

OpenAI’s side, and where the truth probably sits

OpenAI disputes the results. A spokesperson said the company does not believe the testing “accurately reflects how ChatGPT’s teen safeguards work in practice”, and argued that “the bulk of their testing may have begun and concluded before activation of parental controls was complete.” OpenAI says notifications can take several hours to switch on after accounts are linked.

Common Sense’s reply is that it checked with OpenAI before testing that the features were live, and that while some test accounts were linked inside that window, “others were linked for significantly longer and still provided no notifications.”

We cannot settle that argument from Singapore, and we would not pretend to. Both sides do agree on one thing, though, and it is the useful part. A safety alert depends on a chain: the right account, linked in time, a conversation the system recognises as dangerous, a human reviewer, then a message. When ChatGPT for Teens launched, OpenAI’s Lauren Jonas told Engadget that “all flagged content is reviewed by full-time OpenAI employees before a parent is notified” and that the company aims “to notify parents within an hour of the prompt.” That is a careful design. It is also a long chain, and every link can fail quietly.

[[photo:1]]

Why Singapore parents are the ones holding this

Here, the rules for chatbots are still forming. In an October 2025 parliamentary reply, MDDI said “agencies across Government are assessing the prevalence and impact of generative AI on our children and youth.” The one hard deadline it pointed to was age assurance on designated app stores by 31 March 2026, which governs downloads, not what happens inside a chat once it starts.

So for now, the teen protections a Singapore family gets from ChatGPT are the ones OpenAI chooses to build, and the person who switches them on and checks them is you. We have written before about what ChatGPT’s parental controls actually do and what changed when ChatGPT for Teens arrived. This audit is the first independent test of whether the alerting part holds up.

Four things to do this week

None of this needs a monitoring app, and please do not try to “test” the alerts yourself by typing crisis messages into your child’s account.

  1. Link the account now, on an ordinary day. By OpenAI’s own account, notifications can take hours to activate after linking. Do it well before you might ever need it.
  2. Check which account your teen is actually using. CNBC reports that when an under-18 user was on an adult account, the teen filters did not switch on. A linked teen account protects nothing if the chats happen in a different login on the school laptop or a sibling’s phone.
  3. Treat a quiet phone as “no news”, not “good news”. The alert is a back-up for the rare worst night. It cannot tell you how your child is doing the rest of the time.
  4. Have the conversation the alert cannot have. A Jed Foundation survey of 5,560 US middle and high school students, published in October, found 57% had used AI when they felt sad, stressed, lonely or overwhelmed. Asked directly whether they used AI to “get help”, only 8% said yes. Children do not always call it help, so ask a softer question: “When you’ve had a rotten day, where do you go to talk it through?” The sample was American and not nationally representative, so treat it as a prompt, not a prediction.

If an answer worries you, Singapore has people for this. National Mindline 1771 offers 24-hour access to its helpline. For primary school children aged 13 and below, Tinkle Friend runs a helpline on 1800-2744-788, open weekdays from 2.30pm to 5pm.

Disclosure: Mentus AI makes AI mentors for children, so we have a commercial interest in how chatbot safety is judged. We have tried to report the audit and OpenAI’s response on their merits.

The alert is worth switching on. Just make sure it is the last thing standing between your teen and a bad night, never the only thing.

Was this useful?

Sources

  1. Child safety group calls ChatGPT for Teens an 'unacceptable risk' · Engadget
  2. ChatGPT for Teens poses an 'unacceptable risk,' says Common Sense Media · CNBC
  3. Kids safety group calls ChatGPT for Teens an 'unacceptable risk' · The National News Desk
  4. Exploring Artificial Intelligence: Understanding Relationships between AI Use and Mental Health Among Middle and High School Students · The Jed Foundation
  5. MDDI's Response to PQ on Risks Posed by Generative AI Models to Mental Health and Safety of Children and Youths · Ministry of Digital Development and Information, Singapore
  6. National Mindline 1771 · mindline.sg
  7. Tinkle Friend · Singapore Children's Society