OpenAI built a safer ChatGPT, just not for you
OpenAI’s new “teen” version is an admission by the company that it knows its chatbot is harmful and knows how to make it less harmful. It also knows harms are profitable.
SILICON VALLEY, CALIFORNIA (AUGUST 18, 2026) — OpenAI launched a teen version of ChatGPT today. Here’s why the move is an admission not only that the company is cynically serving toxic, harmful features to adults, but that it can switch them off at will and choses not to.
Specifically, OpenAI’s new default experience for users it identifies as being 13 to 17 tightens restrictions on self-harm, violence, eating disorders, dangerous activities, and explicit sexual or graphic content.
The new version also provides study support meant to help students learn rather than simply extract answers or AI-written essays; it guides them toward solutions instead of supplying them.
Overuse and addiction are addressed mainly through tools OpenAI shipped in September 2025, including Quiet Hours, which lets parents block access during set times; today’s addition is Study Hours, which turns Study Mode on by default during periods a teen or parent chooses.
Parents with linked accounts already receive safety notifications in a narrow set of high-risk situations, and OpenAI is now extending those alerts to eating disorders, though the notifications share only limited information, and parents still cannot read their teens’ conversations.
Finally, the teen version is barred from suggesting that it has personal feelings for the user or that it is conscious, and romantic or sexual chats are blocked completely.
Why the teen version of ChatGPT is an admission of guilt
Offering a far less harmful chatbot to teens is a good idea, if it works. (OpenAI places users in the teen experience automatically, either when they state an age between 13 and 17 or when its age-prediction model, which draws on account-level and behavioral signals, estimates that they are under 18; adults wrongly flagged can lift the restrictions by verifying their age, in some countries with a government ID.)
Don’t get me wrong: I am very much against products that raise the risk of suicide, self-harm, brain rot, and addiction, and that foster the delusion that they are conscious, emotional, humanlike entities with which one can form a loving relationship.
The question is: Why does OpenAI continue to sell adults a product that does all this?
The answer is clear: Brain rot, addiction and emotional attachment are profitable for OpenAI.
(The adult version already deflects conversations about suicide and self-harm, and the teen version is only marginally stricter on those topics.)
The admission of moral guilt is this: OpenAI effectively acknowledged today that it knows which features and attributes cause brain rot, addiction, and attachment — and that it knows how to turn them off.
THE NEW THING: My wife and I have been nomads for 20 years now
READ: The Attachment Economy, Computerworld, Gastronomad newsletter, Gastronomad book
LISTEN: Superintelligent, TWiT, Gastronomad podcast
JOIN: The Gastronomad Experience
FOLLOW: Gastronomad on Surf Social, Machine Society on Surf Social, Bluesky, Reddit, Notes, Mastodon, Threads, X, Instagram, Facebook, Linkedin!
LOOK AT: Mike Elgan Photography
It then turned them off for teens and left them on for adults.
Of course, OpenAI is hiding behind the fig leaf of adolescent development — the claim that these features do damage in formative years.
It’s true they do damage to teens But reducing brain rot, addiction, and attachment would also benefit individual users and society at large, and it would in no way diminish ChatGPT’s value as an informational tool.
It would, however, reduce the value of OpenAI. The company has raised some $190 billion, which it must eventually return to investors. It must do that through subscriptions, advertising and other monetization channels.
Brain rot is profitable because, as users are dumbed down and made lazy, they come to depend more on chatbots to function in their jobs and personal lives.
Addiction is profitable because excessive use both strengthens the incentive to subscribe and exposes users to more advertising.
And, of course, attachment is profitable because it is not only an incentive to overuse but also a disincentive to switch to a rival chatbot. (If you’re in love with a chatbot, you won’t switch.) That’s the foundation of the coming Attachment Economy.
Today’s OpenAI announcement is a mixed bag. It’s a good thing to remove harmful chatbot attributes for teens.
It’s a bad thing to keep harmful chatbot attributes for adults.
But most of all, it’s an admission of guilt. OpenAI knows which engineered features of its chatbots are harmful.
For example, the documentation that accompanied GPT-4o’s release two years ago warned that the ability to retain and reuse personal details “creates both a compelling product experience and the potential for over-reliance and dependence.”
Higher daily use of AI chatbots was consistently associated with increased loneliness, greater emotional dependence on AI, more problematic use, and less socializing with real people, according to a joint study by OpenAI and MIT.
In the safety research for the GPT-4o release, OpenAI reported that early testers used language suggesting emotional bonds with the model, including “This is our last day together.” The company warned that users might form social relationships with the AI that reduce their need for human interaction.
OpenAI knows that ChatGPT can create unhealthy attachments in some users. In October 2025, the company published its own definition of the phenomenon, along with its own estimate of how many users exhibit it.
The company had the definition, the measurement, and the fix by October 2025. Ten months later, the safety-tuned configuration shipped to teenagers only.
Why not offer adults the less harmful version? The only conclusion available is that OpenAI weighed user harm against its own profit and chose profit.
