OpenAI began rolling out ChatGPT for Teens on August 18, 2026, routing 13- to 17-year-olds into a tighter version of the chatbot. Parents get real controls. They still don't get the chats.
OpenAI is trying to solve the teen safety problem without turning ChatGPT into a family surveillance tool. That's the pitch, anyway. The company says users identified as under 18 will be moved into the teen experience on free and paid plans, using either a self-reported birthdate or an age-prediction system that looks for signals a user may be a minor. If an adult gets swept into teen mode by mistake, they can verify their age through Persona with a live selfie or government ID. Persona deletes the file within seven days. OpenAI says it never receives the ID or selfie itself.
Once an account lands in the teen experience, the guardrails tighten around self-harm, suicide, eating disorders, graphic violence and sexual content. There are school tools as well. Study Mode is meant to push teens through questions and steps rather than simply hand over finished work, while homework reminders are supposed to nudge them away from copy-pasting assignments. That's useful. It's also the easy part.
The harder choice sits with parents. If a parent links their account to a teen's, they can set Quiet Hours, turn off voice mode, disable memory, block image generation and opt the teen's conversations out of model training. OpenAI's help center also says parents can receive safety notifications when its systems detect serious self-harm concerns or conversations involving violent acts, after review by trained staff.
Here's what they don't get: the conversation itself. OpenAI says parental controls don't let parents or guardians read a teen's chats and don't provide real-time monitoring. A parent can see that a setting is on. They can't read what their teenager typed to the chatbot at 11 p.m.
Nvidia's $40 Billion Spending Spree Revives Dot-Com Bubble Fears Nvidia has spent over $40 billion this year on stakes in OpenAI, SpaceX, Intel, CoreWeave and other companies that also buy its chips. Analysts including Wedbush's Matthew Bryson are drawing direct comparisons to the vendor financing that inflated telecom valuations before the dot-com crash. - nvidia spending 40 billion on company investments - dot com bubble fears nvidia vendor financing
OpenAI rolled out parental controls on September 29, 2025, after the family of Adam Raine, a 16-year-old who died by suicide in April 2025, sued the company over conversations he had with ChatGPT in the months before his death. The age-prediction model followed on January 20, 2026. ChatGPT for Teens is the next step in that same chain, and it shows where OpenAI has drawn the line: more default safety, more parent settings, no full transcript for the household.
The Same Week, Meta Went to Trial #
The timing isn't subtle. On August 18, the same day OpenAI's teen rollout began, a major trial over Meta's treatment of young users opened in federal court in Oakland. A coalition of 29 state attorneys general has accused Meta of designing Instagram and Facebook to hook minors while illegally collecting children's data. AP reported that California, Colorado, Kentucky and New Jersey are leading the trial before Judge Yvonne Gonzalez Rogers, and NBC Bay Area reported that the broader case also names companies including Google, TikTok's ByteDance, Snapchat, YouTube and Alphabet.
Meta denies the claims. Mark Zuckerberg and Instagram chief Adam Mosseri are expected to testify during a trial scheduled to run for weeks, according to current reporting on the case. The states are not arguing over a single bad feature. They are arguing that the business itself was built around keeping young users engaged for longer, even when the company knew the risks.
The OpenAI case is different, but you can see why parents put the two stories together. Meta is fighting allegations that it built an addiction machine. OpenAI is trying to prove it can build a safer AI companion before regulators and courts decide the answer for it.
That doesn't make OpenAI's answer complete. Frankly, a settings panel is not the same thing as knowing what your child is saying to a chatbot in private. The acute-distress alerts matter, and so do default protections around suicide, self-harm and disordered eating. But they are still alerts after the system decides something has crossed a line. Parents worried about the quiet middle, the late-night questions that don't trigger an emergency flag, are left with trust.
The Gap That Matters #
That gap is the real story here. Quiet Hours and blocked image generation are clean controls. Real visibility into conversations would cost OpenAI something: teen trust, usage, maybe the entire pitch that ChatGPT can be a private place to think out loud. The company picked the version of safety that protects the product as well as the user.
If you're a parent linking your kid's account, know what you're actually getting. You can shape the settings. You can receive narrow safety alerts. You cannot read the chats. OpenAI has built a control panel, not a window.
[OpenAI's New Ultrafast Mode Runs GPT-5.6 Sol on Cerebras Chips, Not Nvidia](https://startupfortune.com/openais-new-ultrafast-mode-runs-gpt-56-sol-on-cerebras-chips-not-nvidia/)
OpenAI previewed Ultrafast mode for GPT-5.6 Sol on August 13, hitting 750 tokens per second by running on Cerebras chips instead of Nvidia GPUs. The move builds on a Cerebras deal OpenAI signed in January that the company now values above $20 billion. - OpenAI's GPT-5.6 Sol running on Cerebras chips - Ultrafast mode pushes 750 tokens per second output
Also read: How Does AI Coding Agent Pricing Work, and Where Founders Get Squeezed • Ornn Wants Wall Street to Trade AI Compute Prices Like Oil • TerraPower Breaks Ground in Wyoming to Power the AI Data Center Boom