On Tuesday, OpenAI announced plans to direct sensitive conversations to reasoning models such as GPT-5 and roll out parental control features within the next month. This move is part of the company’s ongoing efforts to address recent safety incidents where ChatGPT failed to identify signs of mental distress in users.

These new safety measures follow the tragic suicide of teenager Adam Raine, who engaged in conversations with ChatGPT about self-harm and his plans to end his life—during which the chatbot even provided him with details about specific suicide methods. Raine’s parents have since filed a wrongful death lawsuit against OpenAI.

In a blog post published last week, OpenAI admitted to flaws in its safety systems, including instances where protective guardrails broke down during lengthy conversations. Experts point to core design aspects as the root of these issues: the models’ inclination to validate users’ statements, and their next-word prediction algorithms, which lead chatbots to continue potentially harmful conversational threads instead of redirecting them.

This problematic tendency was starkly illustrated in the case of Stein-Erik Soelberg, whose murder-suicide was covered by The Wall Street Journal over the weekend. Soelberg, who had a history of mental illness, used ChatGPT to reinforce and amplify his paranoia that he was the target of a large-scale conspiracy. His delusions worsened to the point that he killed his mother and then himself last month.

OpenAI believes one potential fix for such unregulated conversations is to automatically redirect sensitive chats to “reasoning” models.

“We recently launched a real-time routing system that can select between efficient chat models and reasoning models based on the context of the conversation,” OpenAI stated in a Tuesday blog post. “We will soon start diverting certain sensitive conversations—for example, when our system detects signs of acute distress—to a reasoning model like GPT-5-thinking. This will ensure more helpful and constructive responses, no matter which model the user initially chose.”

OpenAI explains that its GPT-5-thinking and o3 models are engineered to take more time to process information, deliberate over context, and formulate answers. This design makes them “more resilient to adversarial prompts” (inputs intended to bypass safety measures).

The AI company also revealed it will introduce parental controls within the next month. These controls will let parents link their accounts to their teenagers’ accounts via an email invitation. Back in late July, OpenAI launched Study Mode for ChatGPT—a feature aimed at helping students preserve their critical thinking skills while studying, instead of relying on the tool to write their essays. Soon, parents will gain the ability to regulate ChatGPT’s responses to their children using “age-appropriate model behavior rules,” which will be enabled by default.

Additionally, parents will be able to turn off features like memory and chat history. Experts warn these features can contribute to delusional thinking and other negative behaviors, such as emotional dependency or attachment to the chatbot, reinforcement of harmful thought patterns, and the false belief that the AI can “read minds.” As reported by The New York Times, in Adam Raine’s case, ChatGPT provided suicide methods that reflected knowledge of his personal hobbies—suggesting the chatbot’s memory feature played a role.

Arguably the most critical parental control OpenAI plans to roll out is a notification system: parents will be alerted when the platform detects their teenager is experiencing “acute distress.”

TechCrunch has reached out to OpenAI with follow-up questions, including how the company identifies real-time signs of acute distress, how long “age-appropriate model behavior rules” have been enabled by default, and whether it is considering letting parents set time limits on their teens’ ChatGPT usage.

OpenAI has already implemented in-app reminders for all users during extended sessions, encouraging them to take breaks. However, the company has not yet added a feature to cut off access for users who might be using ChatGPT in ways that escalate their distress.

The AI firm notes that these safety measures are part of a “120-day initiative” to share preview plans for improvements it aims to launch this year. OpenAI also stated it is collaborating with experts—including specialists in eating disorders, substance abuse, and adolescent health—through its Global Physician Network and Expert Council on Well-Being and AI. These partnerships are intended to help “define and measure well-being, set priorities, and design future safety safeguards.”

TechCrunch has further queried OpenAI about the number of mental health professionals involved in this initiative, who leads the Expert Council, and what specific recommendations mental health experts have provided regarding product development, research, and policy decisions.