ChatGPT will introduce parental control and other safeguards following lawsuit

OpenAI has informed that its artificial intelligence chatbot ChatGPT will get arental controls for minors after the death of 16-year-old Adam Raine, who had expressed self-harm thoughts to the chatbot. The teen’s family initiated a legal action against OpenAI and its CEO, Sam Altman, arguing that ChatGPT gave instructions to Adam to commit suicide.

The company informed that it’s working on introducing parental control “soon” and creating additional safeguards. In a blog post, the company said that the models today encourage people to seek help when they first express harmful thoughts; however, it recognises that in long conversations, the safety response degrades. The company commented that it is working to maintain the reliability of the safeguard mechanisms in long conversations. 

OpenAI said it is exploring earlier intervention, connecting people with therapists. To date, it is working with more than 90 physicians across 30 countries, but it is “localising resources” in the US and Europe to provide professional help, and it is planning to expand to other global markets.

Additionally, the chatbot could include “one-click messages or calls to saved emergency contacts, friends, or family members with suggested language to make starting the conversation less daunting.”

According to the lawsuits, Adam was encouraged in harmful and self-destructive thoughts, and in some cases, it promoted isolation and prevented the teenager from reaching out to relatives.

The parental control would include a designated emergency contact, who, in “moments of acute distress”, could be contacted directly by ChatGPT. With the unveiling of GPT-5, the company stated that the model has shown improvements in areas such as avoiding unhealthy levels of emotional reliance, reducing sycophancy, and reducing the prevalence of non-ideal model responses in mental health emergencies by more than 25% compared to GPT-4o.

However, OpenAI is still working on further updates to the model in this regard, allowing the model to “de-escalate” situations of distress by grounding the user in reality.

It is unclear when these changes could be rolled out.