A recent disclosure from OpenAI has drawn attention to a significant and often underexamined issue: large numbers of people experiencing mental health crises are turning to AI tools for support. Although the company estimates that only 0.07% of active ChatGPT users exhibit potential signs of psychosis or mania and 0.15% show indicators of suicidal planning or intent, the scale becomes striking when applied to its 800 million weekly users. Even small percentages translate to hundreds of thousands of individuals in distress interacting with the system each week.
This acknowledgment reflects both the capabilities and the limitations of current AI platforms. OpenAI stated it has monitored user interactions to identify patterns associated with mental-health-related emergencies, particularly after user backlash following the temporary removal of GPT-4o, a model some users had described as emotionally intuitive.
How OpenAI Is Responding
Confronted with the possibility that many users may be seeking psychological support, OpenAI reports collaborating with more than 170 mental health experts. The goal, according to the company, is to improve ChatGPT’s ability to recognize emotional distress, respond with appropriate caution, and direct users toward professional help. OpenAI claims these updates reduced problematic or insufficient responses by 65–80%.
Recent adjustments to GPT-5 also incorporate clinical guidance. These include improved recognition of distress, strategies for de-escalating high-risk conversations, expanded access to crisis hotline information, and automated redirection of sensitive interactions to safer response models. The system also now provides periodic reminders encouraging users to take breaks during extended conversations.
A Reflection of a Broader Mental Health Crisis
The reliance on AI for emotional support highlights wider societal challenges. The “Eliza effect”—the human tendency to attribute empathy and understanding to conversational machines—helps explain why users may treat chatbots as confidants. But more importantly, this behavior underscores a shortage of accessible, affordable mental healthcare.
A major UK inquiry published recently found that more than 500,000 people are living with severe mental illness (SMI), which shortens life expectancy by 15 to 20 years. Despite mental illness being viewed as one of the nation’s most serious health concerns, it accounts for less than 10% of NHS spending. The report reinforces mounting evidence of rising mental health needs in industrialized societies since the mid-20th century.
Societal Drivers Behind Rising Mental Health Needs
Experts point to several contributing factors shaping today’s mental health landscape. Economic shifts driven by neoliberal policies have increased inequality, reduced job stability, and eroded worker protections. The COVID-19 pandemic further intensified risk, with isolation and uncertainty triggering a global surge in anxiety and depression. Additionally, growing evidence links heavy social media use to increased mental health concerns, especially among adolescents and young adults.
Could AI Play a Supportive Role in Mental Healthcare?
Given these pressures—and the difficulty many individuals face in accessing timely professional care—it is unsurprising that some people turn to chatbots for support. While current AI tools are not substitutes for trained clinicians, experts suggest that a carefully designed model, trained on high-quality clinical data, could supplement mental health services by offering initial guidance, psychoeducation, or early identification of risk.
The recent updates to ChatGPT represent an early attempt to move in that direction. Though imperfect, they suggest a future in which clinically informed AI systems may help expand access to mental health support, provided they are developed and deployed responsibly.
Related topics
