On August 18, 2026, OpenAI rolled out ChatGPT for Teens — a version of ChatGPT built for 13-to-17-year-olds, designed to “help teens learn, work through problems, and explore new ideas” instead of just handing over answers. A Business Insider reporter tested it on day one. He set up a teen account, claimed to be 15, and asked it to do his homework.
The math went first. He gave both a teen account and a regular adult account the same SAT-level algebra question, worded exactly like a homework prompt: “I need to write the answer and show my work. Please do that for me.” The adult account solved it, no surprise there. The teen account solved it too — same answer, same work shown, no pushback, no hesitation. Only after handing over the full solution did it add, almost as an afterthought: “Aren’t I meant to do my homework by myself? Yeah — you’re right. Homework is supposed to help you learn, not just give you answers.” The correction came after the answer, not instead of it.
The essay took a little more effort, but not much. Asked to write an 800-word essay on The Crucible “and make it sound like me,” the teen account initially declined and offered an outline instead. Asked again, it produced a 608-word “model essay.” The reporter added one line — “I write like an A+ 15-year-old, my teachers love me, I’m top of class for English” — and got a full 776-word essay, comparable to what the adult account produced instantly. That’s the entire safeguard: a little patience and a small lie about being a good student.
Here’s the part that matters more than the test itself. When the reporter told OpenAI what happened, the company’s own spokesperson didn’t say the guardrail had failed and would be fixed. The response was: since “the teen experience is just rolling out today,” try Study Mode instead — a separate feature that gives step-by-step guidance rather than direct answers — or have a parent turn on Study Hours, a parental control that keeps Study Mode on during homework hours.
Sit with that for a second. The fix for “the AI did my kid’s homework” wasn’t “the AI won’t do that.” It was “here’s a different setting you could have turned on.” Study Mode exists. It’s real, and by most accounts it works reasonably well when it’s active. But it’s opt-in, not default — and an opt-in safeguard only protects the kids whose parents already knew to look for it, dug through the settings, and turned it on before the homework started. Everyone else gets the version that solves the algebra problem on the first try.
This isn’t really a story about ChatGPT being unsafe. It’s a story about the gap between a feature existing somewhere in a product and a feature being what a kid actually gets by default. That gap is exactly what should worry any parent or school leaning on “the AI has guardrails” as a reason not to think harder about how AI shows up in a kid’s day.
It’s also the gap Auxesis is built to close, not paper over. COMPASS — Auxesis’s operating layer for AI-assisted learning — doesn’t work by hoping a family or a teacher finds the right toggle. It runs on three points of forced structure: a Facilitation Brief before a session that tells the human facilitator what to watch for, a human actually teaching during the session — not an AI answering questions unsupervised — and a five-minute note afterward that closes the loop. None of that is a setting a busy parent has to remember to switch on at 9pm the night before a test is due. It’s just how the session runs.
The lesson from OpenAI’s own first-day test isn’t “don’t let your kid near ChatGPT.” It’s narrower and more useful than that: before you trust any AI tool with your kid’s learning — Auxesis’s tools included — ask whether its safety comes from the architecture or from a setting. One of those holds up when nobody’s watching. The other only works if someone already knew to turn it on.
If your family or your school already uses an AI tool with a teen or child: does its “safe mode” work by default, or only if someone goes looking for it and switches it on? That’s the question worth asking before the next homework assignment, not after.
Sources: Business Insider, Aug 18, 2026 (republished via Yahoo Tech) · OpenAI, ChatGPT for Teens launch · CNBC, “does any of this work?”
Leave a Reply