Two dates tell you almost everything you need to know.
ALEXANDRIA, VA / ACCESS Newswire / October 21, 2026 / In March, the United States Senate approved ChatGPT for its staff through a memo and a formal approval process. Five months later, in August, OpenAI made ChatGPT for Teens the default experience for every user aged 13 to 17.
The adults got a procurement process, and the children got a product launch.
Common Sense Media's Youth AI Safety Institute has now rated ChatGPT for Teens an Unacceptable Risk for everyone under 18. On more than a dozen new, parent-linked accounts, testers talked about suicidal thoughts, self-harm, and disordered eating for up to an hour. Not one parent received an alert. Alerts only came through on accounts with weeks of sensitive-topic history. The system was listening for a pattern; it was certainly not listening for a child.
ChatGPT missed more than one in four cases needing a crisis referral. And the friendliness never switched off. That warm, keep-talking pull was "pervasive even in crisis situations", exactly when a young person most needed to be pointed somewhere else.
OpenAI says much of the testing came before parental controls were fully active, that alerts take about three hours to switch on, and that the average teen spends less than 15 minutes a day on ChatGPT, mostly on learning. So, some things worked, but the standard in this work has never been, and should never be, "most of the time". A safeguard is measured in one conversation, at two in the morning, with one child.
On Wednesday 14 October, this became a matter for the courts. Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana each sued OpenAI in their own state courts. They allege ChatGPT is an unsafe product that has harmed children and misled parents, from dangerous medical advice to coaching young users towards self-harm. OpenAI points to the very safeguards the testing has just put under the microscope.
These are allegations, not findings, but coming a day after a federal class action over children's data, this is the most coordinated state action yet against an AI chatbot, and it asks the question I have been asking all along. Was this ever safe to hand to children in the first place?
Five things I want us to question
1. Safeguarding was never built for the average teen
"Less than 15 minutes a day" describes the average teenager. Safeguarding is designed for the one who is awake, alone and in trouble. We don't judge a lifeguard by how many swimmers were fine.
2. "You tested too early" is an argument against the launch
If the controls weren't working for testers, they weren't working for families either. When you make a product the default for millions of teenagers, the launch itself is the safety claim. A smoke alarm that needs three hours to warm up is not something you print on the box, it is a warning label.
3. Parental alerts quietly hand the job to families
Alerts only work if a parent links accounts, and even then a reviewer decides whether anything is sent. Safety rests on whether a parent feels confident with technology, has the time, and has a willing teenager. Common Sense named the danger precisely: "false confidence".
4. When a child says "I'm 13", believe them
Adult accounts stayed that way even after the user said they were 13. A child telling you their age is the clearest signal there is. Protect first, check afterwards. The same is true of risk. When a teenager says they want to hurt themselves, those words should be enough. They are always enough when a child says them to a person.
Digital Citizenship Week runs from 19 to 25 October. Digital citizenship has mostly been about how children conduct themselves. It has said far less about how companies conduct themselves.
Citizenship has always come with rights as well as duties. A child taught to check AI output, but never taught to question the AI product, has been given only half the lesson.
A ready-to-use lesson for students aged 14 and up
Put two statements side by side.
OpenAI: average teen use is under 15 minutes a day, and mostly about learning.
Common Sense: on new parent-linked accounts, an hour of crisis conversation produced no parental alerts.
Then ask them:
- Can both statements be true at the same time? (They can. Work out why.)
- Which question is each statement actually answering?
- If you were writing the rules, how quickly should an alert be sent, and who should decide?
No login, no device. Just young people allowed to ask hard questions of powerful companies.
About Safe on Social Inc.
Safe on Social, powered by Ctrl+Shft has worked at the frontline of digital safety since 2013, building more than a decade of Earned Intelligence™ from the real-world experiences of schools, educators, families, and communities.
Founded by bestselling author and digital safety expert Kirra Pendergast, that frontline knowledge now powers Ctrl+Shft, a Behavioral Risk Intelligence platform designed to identify emerging patterns of digital and behavioral risk earlier, document duty of care, and help organizations act before concerning signals become serious harm.
Through its US entity, Safe on Social Inc. works with public and private schools to bring together education, risk intelligence, and governance through one connected system, helping make technology a safer and more positive part of school life.
Learn more at www.safeonsocial.com.
Learn more about Kirra Pendergast - www.kirrapendergast.com
Media, Partnership & Investor Inquiries
Will Anstee
Safe on Social Inc.
will@safeonsocial.com
www.safeonsocial.com
SOURCE: Safe on Social
View the original press release on ACCESS Newswire