OpenAI released data showing that more than one million people discuss suicidal thoughts with ChatGPT each week. The firm estimates that 0.15% of all active users have conversations that “contain clear signs of planning or intent to commit suicide.”

Company representatives added that a similar percentage of users show emotional dependence on the chatbot. On top of that, hundreds of thousands show signs of psychosis or mania.

Although such cases are considered rare, OpenAI recognizes their growing social significance.

The new data was published as part of a report on ChatGPT’s security work. The company said it consulted with more than 170 mental health professionals to help improve the model’s responses.

They estimate that GPT-5 responds to crisis topics “more appropriately and consistently” than previous versions.

OpenAI claims that the updated AI model is about 65% more likely to correctly recognize alerts. It is also 91% more likely to meet desired standards of behavior when talking about suicide, compared to 77% for the past version.

The problem became acute after an investigation into the suicide of a teenager who had previously shared suicidal thoughts with ChatGPT. Prosecutors’ offices in California and Delaware are now probing OpenAI and demanding stronger protections for minors.

In response, the company said it is implementing parenting tools and an age prediction system. Its representatives assure that this will automatically identify children and apply strict security measures. OpenAI is also adding new metrics to its basic tests, including emotional dependency and mental health disorders.

The firm’s CEO Sam Altman claims that the company has “been able to mitigate serious mental health issues.” However, he acknowledges that not all model reactions are considered safe.

Despite the introduction of GPT-5, millions of users continue to use older and vulnerable versions of ChatGPT, including GPT-4o.