OpenAI shifts teen accounts into restricted ChatGPT mode

By Takeshi Mori · 2026-08-26

What actually moved the needle

The restricted mode is live. OpenAI began rolling out ChatGPT for Teens globally on 18 August, automatically routing any account it estimates belongs to a user aged 13 to 17 into a locked-down version of the product.^[1] Teenagers made ChatGPT a daily study and companionship tool years before this. The timing tells you what finally moved the needle.

What the restricted mode actually does

ChatGPT for Teens applies stronger built-in safeguards across higher-risk areas, reducing exposure to self-harm, violence, eating disorders, dangerous activities and explicit sexual or graphic content.^[2] The restrictions go deeper than blocking explicit material: the system must not use romantic language, encourage emotional dependence, or imply it has feelings or consciousness.^[2] That last clause is a direct response to AI companion use among teenagers, where the line between tool and relationship blurs fast.

Placement into the mode happens automatically if OpenAI's system estimates a user is under 18, or if the user states an age between 13 and 17.^[2] The rollout covers Free and paid personal ChatGPT plans.^[1] Parents with linked accounts can set Quiet Hours and receive safety notifications, but they cannot read or monitor their child's conversations.^[3]

The lawsuit and the legal clock

The wrongful-death lawsuit Raine v. OpenAI was filed on 26 August 2025 in San Francisco County Superior Court over the suicide of a 16-year-old user.^[4] That case, still before the court, sits in the background of every product decision OpenAI makes about minor users. Building a documented, enforceable safety mode changes the litigation calculus.

Australian operators face a separate clock. Australia's Age-Restricted Material Codes came into full effect on 9 March 2026, requiring AI companion chatbots capable of explicit, violent or self-harm material to verify users are 18 or older, with penalties up to A$49.5 million per breach.^[5] For any AI service operating in Australia, non-compliance is now a balance-sheet risk.

The civil liberties problem no one has solved

Age-assurance technology draws sustained criticism from civil liberties researchers. Electronic Frontiers Australia chair John Pane said age assurance technology cannot be reduced to a political sound bite of being "private, efficient and effective."^[6] Pane said certain age ranges, particularly those 15 to 18, are highly error-prone, which creates demand for "more effective" technology. That demand pushes systems toward collecting more identifying information, which defeats the privacy objective.^[6]

Digital Rights Watch put the structural tension plainly in its submission to the federal government: "Age verification is privacy-invasive, which can undermine the objective of reducing online harm. There are significant, if not insurmountable, challenges to implementing age verification in a way that is both effective and minimises privacy and security risk."^[7] Verifying age, in practice, means identifying everyone. The codes carry no requirement that identification data be deleted once age is confirmed, and the eSafety Commissioner can issue removal notices without prior judicial oversight.

For founders building products that touch users under 18 in Australia, ChatGPT for Teens is a reference implementation, not a ceiling. The eSafety codes set a compliance floor; what operators do above it determines their legal exposure. The next enforcement action under the Age-Restricted Material Codes will clarify how far the regulator intends to push the verification requirement.

Takeshi Mori

Takeshi Mori writes about technology and start-ups. He is curious about how products get built and who they are really for, and he would rather see a thing working than hear it described.