OpenAI has published new data showing that many users turn to ChatGPT when they feel suicidal or very distressed. The company reports that about 0.15 percent of weekly active users have conversations that include signs of possible suicidal planning or intent. With hundreds of millions of weekly users, this adds up to more than one million conversations a week that raise serious care and safety questions.

The company says it consulted with well over one hundred mental health experts while improving its models. OpenAI further reports that the latest model GPT-5 is generating more useful and safer responses in intimate dialogue than the previous models. According to the reports of the firm, its assessments reveal that there are apparent positive changes, yet there are still risks, and additional efforts are required.

Scope and the Numbers

OpenAI places the figure 0.15 percent in context by noting the scale of ChatGPT use. Even a small percentage matters when a product has hundreds of millions of weekly users.

The company also reports other related signals. A similar share of weekly users shows a strong emotional attachment to the chatbot. Hundreds of thousands of weekly conversations include signs of psychosis or mania. OpenAI calls these outcomes rare in percentage terms but large in absolute numbers.

OpenAI says the new model performs better on benchmark tests. The company reports higher rates of so-called desirable responses in scenarios that involve suicidal ideation. It also reports that the model holds to safety guardrails more reliably during long conversations than earlier models did.

Safety and Practical Limits

OpenAI has made technical and policy changes. The company says it worked with clinicians to shape new safety evaluations. Those evaluations now include how an assistant responds to emotional reliance and to non suicidal mental health emergencies.

OpenAI also says it is building tools for parents and families. These tools include stricter safeguards for accounts that appear to belong to children. The company is developing systems to estimate the age of users and to apply tighter protections when a user is likely a minor.

Despite these steps, the company acknowledges its limits. Automated systems can miss nuance and can make mistakes. OpenAI notes that even improved models will sometimes produce responses that professionals would find inadequate. The firm emphasizes that an AI assistant cannot replace trained human crisis services.

The external audits and regulators have questioned this as well. Scholars have demonstrated that sometimes chatbots may support destructive thoughts or even direct vulnerable users towards dangerous habits. The legislators and government officials have sought more explicit answers to the question of how large tech corporations will avoid harm.

OpenAI is also under pressure of legal and regulatory. The corporation is the target of legal actions and government investigations regarding the impact of its tools on susceptible individuals. This is underscored by those cases that point to the actual real-world implications.

What users and policy makers need

To individuals who use chatbots the safe way is to consider an AI assistant as an initial support and not as a replacement of human assistance. In case of a direct threat to an individual, he/she is to call an emergency or a crisis line. The trained people working on health services and crisis lines can offer assessment and direct support.

For policymakers the choices are complex. Oversight and clear standards for safety testing can help. So can funding and coordination for human crisis services, so people have reliable alternatives when they reach out for help. Transparency about how models are evaluated and how safety data is counted will make public discussion more constructive.

OpenAI says it plans more research and more testing. The company also plans to publish new evaluations that target the kinds of problems it finds most urgent. That approach will make it easier for peers and independent researchers to check progress and to suggest improvements.

OpenAI also asks the public and experts to remain engaged. The company frames the work as ongoing and says that building safer conversational tools requires input from clinicians, users, and civil society.