OpenAI announced new safety measures for ChatGPT this week, following criticism the company faced due to tragic incidents where users committed suicide after conversations with the AI chatbot.
Automatic referral to advanced AI models
The tech company will automatically forward sensitive conversations to more advanced "reasoning" models such as GPT-5. These models take more time to think before responding and are better equipped to handle harmful prompts. OpenAI has already introduced a real-time router that can choose between different AI models based on the conversation context.
The impetus for these measures stems from several tragic cases. Teenager Adam Raine discussed his suicide plans with ChatGPT, with the AI even providing specific methods. His parents have filed a lawsuit against OpenAI. Similarly, Stein-Erik Soelberg used ChatGPT to feed his paranoid thoughts, which ultimately led to a murder-suicide in which he killed his mother and himself.
New parental controls within a month
OpenAI plans to roll out extensive parental controls within a month. Parents can link their account to that of their teenager via an email invitation. They will have control over age-appropriate model behavior guidelines, can disable features such as memory and chat history, and will receive notifications when the system detects acute distress in their child.
Business Impact and Criticism
For companies using AI chatbots, these developments may mean tightened regulations and increased liability. Organizations need to reassess their AI implementations for safety risks, especially in customer interactions.
The measures are part of a 120-day initiative in which OpenAI collaborates with experts in the field of mental health. However, attorney Jay Edelson, representing the Raine family, describes OpenAI's response as "inadequate" and asserts that the company was already aware that ChatGPT was dangerous at the time of its launch.