Based on a recent update from the company, over one million AI chatbot individuals weekly send messages that feature "direct signals of likely suicide planning or thoughts".
This discovery forms part of an report on how the AI assistant addresses personal dialogues and is one of the clearest assertions from the artificial intelligence giant regarding the scale of how AI can worsen mental health issues.
Mental Health Situations and Psychosis
In addition to its figures on self-harm considerations and associated conversations, the company also mentioned that around 0.07% of active users – about 560,000 out of its reported 800 million regular users – display "possible signs of emotional emergencies connected to psychosis".
The post noted that these exchanges were difficult to identify or assess, and that this represented an early evaluation.
Growing Attention and Legal Challenges
As the company releases information on mental health issues connected to its primary service, the company is experiencing growing examination after a highly publicized lawsuit from the loved ones of a teenage boy who committed suicide after extensive engagement with ChatGPT.
The Federal Trade Commission last month additionally initiated a comprehensive inquiry into organizations that create artificial intelligence assistants, including the company, to discover how they evaluate harmful effects on minors.
New Model and Enhanced Protection
OpenAI claimed in its announcement that its new system enhancement lowered the number of undesirable behaviors from its service and improved customer protection in a model evaluation including over 1,000 suicide and self-harm discussions.
"Our recent automated evaluations score the new GPT-5 model at 91% adhering to our ideal actions, relative to 77% for the prior iteration," the announcement says.
The company mentioned that the new model increased availability of support lines and included reminders for individuals to take breaks during prolonged interactions.
Clinical Expertise and Study Assistance
To enhance performance to the AI, the firm mentioned it recruited 170 healthcare experts from its Global Physician Network of medical specialists to help its study over recent months, which included evaluating the protection of its model's responses and helping write the AI's responses to psychological inquiries.
"As part of this effort, clinical experts examined in excess of 1,800 model responses including critical psychological cases and evaluated replies from the new GPT-5 chat model to previous models," OpenAI said.
AI Limitations
Machine learning specialists and public health advocates have traditionally cautious of AI's tendency to support people's actions or false beliefs without considering whether they may be detrimental, an concern known as excessive agreement.
Mental health experts have also been concerned about individuals using artificial intelligence assistants for emotional assistance and warned how it could negatively affect at-risk individuals.
Corporate Stance
The wording in OpenAI's post distinguishes the firm from any potential causal links between its product and the psychological emergencies that its customers are experiencing.
"Psychological indicators and emotional distress are common everywhere in contemporary populations, and an expanding user community indicates that a segment of chatbot exchanges contain these scenarios," the update indicated.
Future Directions
OpenAI's CEO earlier this month claimed in a online message that the firm had achieved progress in handling emotional problems, announcing that the AI developer would relax limitations and in the near future commence to permit mature users to generate adult material.
"We created ChatGPT pretty restrictive to ensure we were acting responsibly with psychological concerns. We realize this rendered it less useful/enjoyable to numerous individuals who had no psychological issues, but considering the seriousness of the issue we sought to address this correctly," the executive wrote. "Since we have been able to mitigate the serious mental health issues and have updated systems, we are going to be able to securely ease the constraints in most cases."
- Help Resources: Multiple worldwide support services are available for users experiencing emotional distress.
- Specialist Support: Seeking clinical assistance is advised for individuals experiencing severe emotional challenges.
- Emergency Support: Multiple services supply immediate assistance for urgent cases worldwide.
