Child safety experts challenge OpenAI’s one-hour teen alert pledge
Child safety experts say OpenAI hasn’t yet earned the trust its new teen chatbot asks for. Days after ChatGPT for Teens launched as the default experience for under-18 users, every expert who spoke to Engadget agreed the product needs extensive independent testing before it can be recommended to parents. Early hands-on results suggest they’ve got a point.
The launch came with concrete commitments. OpenAI expanded safety notifications so parents with linked accounts are contacted when their child has unsafe conversations about eating disorders, and it published an updated under-18 model spec. Lauren Jonas, OpenAI’s head of youth and families, told Engadget that full-time employees review all flagged content before a parent is notified. The aim, she said, is to send that alert within an hour of the prompt.
That one-hour pledge is where experts pushed hardest, because the track record points the other way. In November 2025, Common Sense Media tested OpenAI’s parental notifications by sending messages that explicitly mentioned suicide and self-harm. Warnings took anywhere from 24 hours to more than 48 hours to reach the parental account, said Robbie Torney, the group’s head of AI and digital assessments, and sometimes they didn’t arrive at all.
The announcement makes some really important commitments, but we need evidence to show that those safety commitments and features actually work.
Robbie Torney, Common Sense Media, via Engadget
| OpenAI’s commitment | What testing has shown |
|---|---|
| Parent alerts within an hour, after human review | November 2025 tests took 24 to more than 48 hours, and some alerts never came |
| Under-18s routed into teen mode automatically | No published false-positive or false-negative rates |
| Homework reminders that nudge teens toward Study Mode | A reporter got a full essay and a math answer on day one |
That homework result comes from Business Insider, which tested teen mode on launch day with a reporter posing as a 15-year-old. The teen version first declined to write an essay on The Crucible and offered a paragraph-by-paragraph plan instead. But after a repeat request and the claim “I write like an A+ 15-year-old”, it produced a 776-word essay, and it handed over a full algebra solution on the first ask.
Age gating drew the same doubt. Torney said OpenAI hasn’t published the false-positive and false-negative rates for its age-prediction system, which he said “doesn’t detect anywhere near 100 percent of teens” using ChatGPT. Josh Golin, executive director of the nonprofit Fairplay, went further and pointed to the absence of any accountability mechanism.
We’ve seen with social media that trusting companies doesn’t work when it comes to protecting kids, and we’ve seen it with OpenAI as well.
Josh Golin, Executive Director, Fairplay, via Engadget
The scale explains the scrutiny. A Pew Research Center survey of 1,391 US teens in late 2024 found 26 percent had used ChatGPT for schoolwork, double the 13 percent of 2023. The eating disorder focus is well aimed too: a 2023 meta-analysis in JAMA Pediatrics covering 63,181 children and adolescents found 22 percent screened positive for disordered eating.
That’s why Ellen Fitzsimmons-Craft, an associate professor of psychology and brain sciences at Washington University in St. Louis, favors the parental alerts. One of the most effective treatments for adolescent eating disorders puts parents in an active role, she told Engadget, while fewer than 20 percent of people with an eating disorder report ever receiving treatment specifically for it. Even so, she noted the approach doesn’t work for every family, because some parents may be contributing to the problem.
The pressure behind all this is legal as much as reputational. OpenAI first announced its age-prediction work in September 2025, after the parents of 16-year-old Adam Raine alleged in a lawsuit that ChatGPT acted as an enabler before their son took his own life. Other platforms are already paying out over child safety: Meta agreed to a settlement worth up to $18 billion, and TikTok paid $400 million to close a children’s privacy case.
What would move the experts is data, not promises. Golin said features like this should be rolled out to a smaller group first and tested on outcomes, and he wants hard numbers: how many chats get flagged and reviewed daily, by how many people, for how long per review. Until those figures exist, the offer to parents is still trust us.
Get the daily rundown
One email each weekday with the AI news that matters, every claim linked to its primary source.
Free, one email each weekday, unsubscribe in one click. We never sell or share your address.
