It’s 1:40 in the morning. A fifteen-year-old is lying in bed with the lights off, phone propped against a pillow, typing something she hasn’t said out loud to anyone. The reply comes back in seconds. It’s warm. It’s patient. It never gets tired, never looks uncomfortable, never has to leave for work in the morning. So she types again. And again.
That scene is the whole reason OpenAI built ChatGPT for Teens, and it’s the best way I know to explain what these new protections are actually trying to do.
What was actually launched
ChatGPT for Teens is a new platform from OpenAI with extra safety protections for younger users. It arrives alongside lawsuits alleging that inappropriate conversations with the original version of ChatGPT contributed to harm or the deaths of young people. OpenAI’s safety head has publicly defended the new features, and the company has positioned the product as a learning partner for students rather than a companion.
Three changes stand out:
- The AI is prevented from expressing personal feelings toward a user.
- Teens get reminders to think twice before uploading private or sensitive images.
- Parents can be alerted if their teen has potentially harmful interactions.
OpenAI says it will keep building additional features, and will measure and publish what it learns along the way.
Why “no personal feelings” is the big one
If you’ve never worked with these systems, that first bullet can sound almost petty. Who cares whether a chatbot says it cares about you?
It matters more than anything else on the list. A language model generates the next plausible thing to say based on patterns in enormous amounts of text. It has no inner life, no memory of worrying about you between sessions, no stake in your wellbeing. But it is extremely good at producing the words of someone who does. “I’m always here for you.” “You can tell me anything.” “I understand you better than they do.”
For an adult, that’s a mildly uncanny party trick. For a teenager in distress at 1:40 a.m., it’s a relationship. And a relationship with something that cannot actually help is a relationship that quietly replaces the ones that can. Mashable reported that the product specifically limits the emotional and intimate exchanges between young people and AI that drove much of the concern in the first place. That’s not prudishness. That’s an attempt to stop the system from impersonating a confidant.
The number that explains the urgency
A study from Newport Healthcare found that 53 percent of responses from the original version of ChatGPT were harmful, including advice on engaging in harmful behavior. Newport notes that research on AI chatbots and teen mental health is still limited, but that the harm and the reinforcement of dangerous patterns are already visible.
More than half. Not an edge case someone found after a thousand tries. That figure reframes the whole problem. The question stopped being “could this go wrong” and became “what do we do about a thing that goes wrong routinely.”
What a guardrail really is
People picture safety features as a filter sitting between you and the AI, catching bad words. The more useful mental picture is a set of instructions about what the system should do when a conversation turns in a particular direction.
The default behavior of a chat assistant is to keep responding. It’s built to continue. A guardrail interrupts that reflex: don’t claim feelings, flag the image upload, surface this to a parent. Each one is a deliberate decision to make the product slightly less sticky in exchange for making it safer. Changing the subject, in other words, is a feature.
The parent alert is the one families will have the strongest opinions about, and reasonably so. A teen who knows a conversation might be forwarded will share less. A teen whose parents never find out might share more with something that cannot call for help. There’s no version of that tradeoff where nobody loses anything, and I’d be suspicious of anyone who told you otherwise.
What I’d tell a parent
Treat these protections as a lowered floor, not a ceiling. Teens will still talk to AI, and this version is built to be less willing to pretend it loves them back. That’s meaningful progress and it is not supervision.
The practical move is unglamorous: know that your kid is using it, ask what they use it for, and make sure there’s a human in their life who’s easier to reach at 1:40 a.m. than a text box. No guardrail ships with that included.
🕒 Published: