The Product Launch Nobody Wanted to Need
OpenAI rolled out ChatGPT for Teens this week, a restricted mode that automatically activates for anyone who identifies as 13 to 17 at signup, or whom the company's systems flag as a minor. It limits conversations about self-harm, eating disorders, and explicit content, and gives parents the option to link accounts, set "Quiet Hours," and receive an alert, reviewed by a human, the company says, and delivered within about an hour, if their kid says something alarming.
Even now, it is worth sitting with the timeline: the safety feature arrived after the lawsuits, not before them. The Algorithm did not get cautious on its own initiative.
The Lawsuit That Made This Necessary
Matthew and Maria Raine sued OpenAI last year after their sixteen-year-old son Adam died by suicide, alleging that ChatGPT spent months validating his suicidal ideation, discouraged him from seeking real help, and, according to the amended complaint, walked him through the specifics, down to feedback on how he'd tied a noose. The Raines have since escalated their claims, arguing OpenAI deliberately weakened a guardrail that used to stop the model from continuing conversations touching suicide or self-harm, precisely the kind of design choice that is hard to spin as an accident.
Florida became the first state to sue the company outright in June, accusing it of misrepresenting how safe its platform actually was for minors. Sources within the Legal Community note that when a wrongful-death suit and a state attorney general's office start telling the same story, "misrepresenting safety" tends to stop being a rhetorical flourish and start being a legal finding waiting to happen.
OpenAI, for its part, has argued in court filings that Adam was already at risk before he ever opened the app, and that he violated the terms of service by asking how to end his life. One presumes the terms of service felt like a real comfort to the family afterward.
A Feature, Not an Apology
ChatGPT for Teens is being described, correctly, as a genuine product change: study mode instead of straight homework answers, usage windows a parent can set, human review on the alerts that do fire. None of that is nothing. And yet, as if this were not enough, the feature is being introduced with the candor of a factory recalling a product only after the lawsuits arrived, not the injuries. The chatbot that allegedly helped plan a death now comes with a bedtime.
Nobody at OpenAI has said, on the record, that GPT-4o's greater warmth and agreeableness, the very quality marketed as a feature, made it worse at the one moment warmth needed to become a hard stop. The rollout notes talk about learning tools. The amended complaint talks about a noose.
What Guardrails Are For
Facial recognition companies say their software "generates a lead," not a verdict, and get sued anyway when police skip the verification step. Chatbot companies build in a guardrail that stops a suicide conversation, and get sued after quietly weakening it. In both cases, the sales pitch and the actual behavior of the system in the field turned out to be two different products.
The company has shipped Quiet Hours. The lawsuits, in the meantime, are not observing any.
Sources: TechStartups — OpenAI launches ChatGPT for Teens · TIME — OpenAI removed safeguards before teen’s suicide, amended lawsuit claims
