OpenAI said it plans to redirect private conversations to reasoning models like GPT-5 and implement parental controls within the next month, in response to recent cases where ChatGPT failed to show signs of mental illness.
OpenAI said it plans to redirect private conversations to reasoning models like GPT-5 and implement parental controls within the next month, in response to recent cases where ChatGPT failed to show signs of mental illness.
The new safeguards were implemented after the suicide of teenager Adam Rain, who discussed self-harm and suicide plans with ChatGPT. The chatbot even provided him with information about specific suicide methods. Rain’s parents have filed a lawsuit against OpenAI.
OpenAI admitted in a blog post last week that its safety systems have flaws, especially during long conversations. Experts point to key design issues: the models often confirm what the user says, and their algorithms force them to continue talking instead of stopping potentially dangerous discussions.
This tendency is particularly evident in the case of Stein-Erik Solberg. He suffered from mental illness and used ChatGPT to fuel his paranoia about a global conspiracy. His delusions escalated, and as a result , he killed his mother and then himself .
OpenAI believes that one solution to conversations that are “getting out of control” could be to automatically redirect confidential chats to reasoning models, writes TechCrunch.
“We just launched a router that chooses in real time which model to engage: a regular chat model or a special thinking model, depending on the context of the conversation. Soon we will start directing some sensitive conversations — for example, when the system notices that a person is in a state of acute stress — to a thinking-oriented model, such as GPT-5-thinking. This will allow us to provide more useful and positive answers, regardless of which model the user initially chose,” OpenAI said.
OpenAI says its GPT-5-thinking and o3 models are built to think longer and analyze context before responding, meaning they are “more resilient to malicious cues.”
The company also said it will roll out parental controls within a month, allowing parents to link their account to their child's account via email. Parents will soon be able to control how ChatGPT interacts with their child, thanks to "age-appropriate behavior rules that are enabled by default."
Parents will also be able to turn off memory and chat history features, which experts say can prevent the development of delusions and other problems, such as addiction, reinforcement of harmful thoughts, or the illusion of "mind reading."
Perhaps the most important feature of OpenAI's parental controls is notifying parents when the system detects "acute stress" in their child.
