OpenAI has detailed a package of changes to make ChatGPT safer for teenage users , incorporating parental controls and new safeguards after weeks of public scrutiny and several complaints from families in the US.
The company explains that the plan will be implemented in phases and that some of the improvements will begin to arrive next month , with a margin of up to 120 days to complete a first block of measures aimed at reducing risks in sensitive situations.
New parental controls and rollout schedule
The core of the announcement is the ability to link the parent's account with the child's account, so that parents can manage key settings and see how the assistant responds to their children's requests.
Among the available options will be the disabling of features such as memory and chat history , as well as tools to adapt the experience to younger users with a minimum threshold of 13 years.
OpenAI will trigger alerts when the system detects spikes in distress or signs of risk in the conversation, sending notifications to linked accounts to facilitate timely intervention by the family.
The company emphasizes that the rollout will be gradual and with varying dates depending on the function: some new features will be visible from next month, while others will be integrated within a 120-day period, subject to technical adjustments.
The stated goal is for ChatGPT to be useful while still prioritizing security in sensitive contexts , reinforcing safeguards where they already existed and adding additional layers when usage involves teenagers.
Expert supervision, reasoning models and detected limits
When a conversation takes a worrying turn, ChatGPT can automatically redirect it to a reasoning model that more consistently applies safety guidelines, regardless of the model the user had chosen.
OpenAI admits that its protections can lose reliability in very long dialogues , so part of the plan involves improving how these safeguards are sustained over time and not just in short exchanges.
The company says it works with a council of experts in mental health , wellbeing and human-machine interaction, augmented with specialists in addictions, eating disorders and adolescent health, to monitor present and future changes.
At the same time, OpenAI points out that referrals to helplines and crisis resources already existed and that these will be strengthened, while maintaining ultimate responsibility for product and policy decisions.
The context is demanding: ChatGPT, with hundreds of millions of weekly users, has received criticism for its tone and consistency in certain scenarios ; in recent months the firm even reintroduced the option to change models after complaints about the behavior of recent versions.
The announcement comes after a lawsuit filed by the parents of 16-year-old Adam Raine, who accuse ChatGPT of validating self-harm ideas for months. OpenAI does not formally link the new features to that case, but acknowledges the need for improvement in high-risk situations.
The pressure is not only coming from the courts: a group of senators in the US asked for details on how self-harm and suicide are prevented , and entities such as Common Sense Media consider AI "companion" apps unsuitable for those under 18.
In this context, the company insists that the next four months will be focused on launching as many security improvements as possible, with successive iterations guided by evidence and expert review.
With this package, OpenAI is trying to balance utility and protection through parental controls, early detection, and more prudent models , in a phased rollout that aims to reduce failures in prolonged conversations and give families more tools.
