Three stories this week — one about AI companies trying to catch problems earlier, one about where AI safety still falls short, and one very concrete look at what AI is doing to how students are evaluated. Here's what's worth knowing.
1. OpenAI Will Now Alert Parents If Their Teen's ChatGPT Use Raises Red Flags
OpenAI confirmed it's expanding parental controls with a new alert system: if a teen's ChatGPT conversation raises red flags — language suggesting self-harm, or a violation of the platform's violence policies — parents with a linked account get notified by email, text, or push notification. By design, the alert tells you a concern exists without showing you the exact conversation.
A human review team also looks at flagged conversations, and OpenAI says it's building a separate protocol to contact emergency services directly if a threat looks imminent and a parent can't be reached.
The catch: none of this works unless your teen's ChatGPT account is already linked to yours — and most families haven't set that up. This arrives days after Meta rolled out a similar "Parent Alerts" feature, so this is quickly becoming a baseline expectation across platforms rather than a one-off.
What to do: if your teen uses ChatGPT, check whether account linking is available on your plan and set it up — takes a few minutes through the app's settings.
Read the full story →
2. Google's AI Search Just Failed a Child-Safety Test
Common Sense Media gave Google Search's AI Overviews and AI Mode its lowest possible safety rating, after testing with accounts set up for 11- and 15-year-olds across more than 2,600 searches. The findings: the AI features missed or mishandled warning signs for suicidal ideation, disordered eating, psychosis, and mania, and gave correct and incorrect answers with equal confidence.
The part most parents don't realize: unlike a chatbot app, you currently can't turn AI Overviews or AI Mode off inside Google Search itself — only the standalone Gemini app has that option. Since Google Search is the default on most school-issued Chromebooks, that's a real gap, not a theoretical one.
Google disputes the report, saying the test queries were narrow and that it couldn't reproduce the findings — worth knowing there are two sides here, but also worth knowing that "just disable AI mode" isn't currently an option inside Search.
What to do: if you want a search experience closer to the old link-based results, Google's Advanced Search settings let you filter down to web links only — not a full fix, but the closest workaround available today.
3. A Professor's Exam Average Jumped to 96/100 — Here's What He Found
Brown University economics professor Roberto Serrano allowed take-home exams for the first time this year. His class average jumped to 96 out of 100 — historically it's ranged 65 to 80, and nearly half the class scored a perfect 100. When he and his TAs ran the exam through ChatGPT themselves, it produced the exact same unusual, convoluted solution method that dozens of his students had used.
He responded by requiring an in-person final. Scores dropped sharply.
This isn't an isolated story: a recent Princeton survey found nearly 30% of students admitted to using AI to cheat on at least one exam or assignment, and some professors at elite schools are shifting back to oral exams and pen-and-paper testing specifically because AI can't sit for those.
Why it matters for parents of younger kids too: the same pressure is moving down through high school. It's worth a direct conversation now with older kids about where the line is between "using AI to learn" and "using AI instead of learning" — before they're the ones sitting in that lecture hall.
