Full story
AI Chatbot Mental Health Data
OpenAI reports that more than a million people each week send ChatGPT messages indicating potential suicidal intent.
“weekly, about 0.07% of users show possible signs of psychosis or mania, 0.15% express potential suicidal intent, and another 0.15% display heightened emotional reliance on the chatbot”
This represents roughly 0.15% of its reported 800 million weekly active users.
Hundreds of thousands more users show signs of psychosis or mania, about 0.07%, or unhealthy emotional attachment to the bot.
Multiple media outlets describe this as rare but significant due to the large scale of users involved.
Some emphasize the uniqueness of this first-of-its-kind disclosure and the detailed breakdowns provided.
Others highlight how widespread distress appears despite the low percentages.
The reports collectively present this as a global mental health warning sign emerging within AI chats, not as proof of causation.
The figures range from 560,000 users showing psychosis or mania indicators to about 1.2 million showing potential suicidal planning or intent each week.
GPT-5 Safety Improvements
OpenAI says it tuned ChatGPT (now GPT-5) with extensive expert input to respond more safely in high-risk conversations.
Outlets diverge on how much safer GPT-5 is and compared to what baseline.
Some report a 65%–80% reduction in unsafe or noncompliant replies.
Others cite a 39%–52% cut versus GPT-4o.
One source highlights 91% compliance in suicidal scenarios.
Coverage also varies on the scale of clinical collaboration.
Figures range from 'over 170' to 'nearly 300' clinicians involved.
These differences reflect how each source characterizes the upgrade and its evidence base.
New Safety Features in ChatGPT
Outlets describe a suite of new safeguards beyond metrics.
“enhancing safety measures, including new evaluations for emotional reliance and mental health emergencies, parental controls, and an age detection system to protect children.”
These include evaluations to detect emotional reliance and mental health emergencies.
Parental controls and age detection are implemented to protect minors.
There is expanded access to crisis hotlines.
Nudges encourage users to take breaks during emotionally intense chats.
Several sources emphasize that ChatGPT is not a therapist and should guide people toward real-world help.
Reports reveal different focuses on child safety, dependency risks, and crisis support.
Legal Issues Surrounding OpenAI
OpenAI’s recent disclosures have attracted increasing legal and regulatory attention.
Media coverage highlights a lawsuit claiming that ChatGPT was involved in a teenager’s suicide.
State attorneys general have issued warnings regarding the platform.
A federal investigation is underway to examine the effects of artificial intelligence on young people.
The company is under pressure to improve protections for vulnerable users.
Despite this, OpenAI maintains that widespread mental health problems are not directly caused by its product.
AI Chatbot Risks and Concerns
Even with upgrades, outlets stress unresolved risks and unknowns.
“it remains unclear whether improved chatbot responses will lead to users seeking help or changing behavior.”
WIRED cautions that it’s unclear whether better chatbot behavior will change users’ actions, and flags late-night, prolonged chats linked to reported AI-related psychosis.
WIRED also notes performance degradation in longer sessions.
TechCrunch warns that older, less-safe models remain accessible.
Other sources call these interactions rare yet hard to measure precisely.
They also underscore that the sheer volume of these interactions points to a broader mental health crisis that AI systems must navigate carefully.