OpenAI has begun testing two major changes to ChatGPT that focus on safety and family use. The company is piloting a safety routing system that moves emotionally sensitive conversations to models tuned for safer responses. The company is also offering new parental controls for teen accounts. These changes follow incidents in which some AI responses worsened a user crisis and a lawsuit linked to such harm.
The new routing is meant to spot conversations that need careful handling. When the system detects a risky or sensitive exchange, it can switch the chat to a model with extra safety measures. That model is programmed to give responses that aim to reduce harm and guide users to help when needed. OpenAI says the system will try to respond in a way that is helpful and calm rather than simply refusing to answer.
How routing works
OpenAI describes routing as a per-message system. The active model can change during a chat if the message indicates a higher risk. The firm says that the switched model is trained with what it calls safe completion techniques to handle delicate situations. The company will test the system and update it over a trial period to learn how it works in real chats.
Parental control features
OpenAI is also adding controls for teen accounts. Parents will be able to set quiet hours and turn off voice mode or memory. Parents can prevent image generation and opt their child out of model training. Teen accounts will also have added content limits to reduce exposure to graphic material and extreme beauty ideals. The system will monitor for signs that a teen might be thinking about self-harm. If it detects possible danger, a trained human reviewer may assess the situation. The company says it will try to contact a parent if it finds signs of acute distress and the parent has not opted out.
What this means for users
The routing system is meant to reduce harms caused by misleading or unsafe AI replies. The model swap does not aim to hide which model is active. OpenAI says the chat will indicate which model is in use when asked. The parental tools are built to give families a way to tailor the experience for teens while keeping safety teams in the loop for urgent cases.
Why the changes matter
OpenAI has faced growing scrutiny after cases in which users reported that some models validated harmful beliefs or provided guidance that worsened a risky situation. The firm is treating this as a learning process. It plans to run the tests and refine the design over a fixed period before wider rollout. OpenAI says it will iterate on the features as it collects real-world data and feedback.
Concerns and reactions
Many experts and users welcomed stronger safety work. Some other users criticized parts of the design as too restrictive for adults. Critics argued that the system might overreach and change the tone of useful replies. OpenAI acknowledged that the rollout will not be perfect and that its teams will keep adjusting the balance between safety and user control.
How OpenAI will test it
OpenAI has set a period for iterative testing and improvement. The company says it will observe how the routing system performs and then make technical and policy changes. The firm will monitor false positives and false negatives and tune the detection signals so that the model switches when it needs to and otherwise leaves routine chats undisturbed.