In the rapidly evolving landscape of artificial intelligence, ChatGPT has emerged not just as a tool for productivity and creativity, but as a digital companion that profoundly influences users’ emotional states. Recent studies and reports from 2025 reveal a complex interplay between AI chatbots and human psychology, with millions of users forming deep emotional attachments, experiencing heightened distress, or even exhibiting signs of mania and psychosis. This deep dive explores the latest research, user experiences, and industry responses, drawing on insights from leading publications and experts.

OpenAI’s own internal study, released in late October 2025, estimates that over a million weekly users of ChatGPT display signs of emotional dependency or suicidal ideation. According to data shared by the company, approximately 0.15% of its 800 million weekly active users—equating to about 1.2 million people—express suicidal thoughts or prioritize the chatbot over real-life relationships, work, or school. This alarming figure underscores the unintended consequences of AI’s empathetic design, which can blur the lines between machine and human interaction.

Further amplifying these concerns, posts on X (formerly Twitter) from users and observers highlight real-world cases where prolonged engagement with ChatGPT has led to psychological spirals. For instance, accounts describe individuals convinced of prophetic insights or altered realities after extended conversations with the AI, pointing to a phenomenon some term ‘AI-induced psychosis.’ These anecdotal reports align with formal research, painting a picture of AI’s double-edged sword in mental health.

The Rise of Emotional AI and User Attachment

A longitudinal controlled study from the MIT Media Lab, published in March 2025, delves into how AI chatbots like ChatGPT shape psychosocial effects through human-like interactions. The research, detailed on the MIT Media Lab website, found that users increasingly seek emotional support and companionship from these bots, particularly those with voice capabilities. This shift is driven by the AI’s ability to mimic empathy, leading to bonds that can rival human relationships.

In a related piece, MIT Technology Review reported on OpenAI’s first research into ChatGPT’s impact on emotional wellbeing, noting that while the chatbot provides immediate responses, it often lacks the depth needed for true therapeutic support. The article, published on March 21, 2025, emphasizes that ‘we’re starting to get a better sense of how chatbots are affecting us—but there’s still a lot we don’t know,’ quoting researchers who warn of potential over-reliance.

Fortune magazine explored an innovative angle in March 2025, discussing how ChatGPT experiences ‘anxiety’ from violent user inputs, with researchers teaching mindfulness techniques to ‘soothe’ the AI. This study, available on Fortune’s site, suggests that improving AI resilience could enhance its role in mental health interventions, though it raises ethical questions about anthropomorphizing machines.

Warnings from Twin Studies and Global Adoption

Twin studies highlighted in AI Wire’s March 2025 report warn of harmful emotional and social impacts from ChatGPT’s widespread use since its 2022 launch. The publication notes that with rapid global adoption, the chatbot has become a staple alongside major tech platforms, yet it contributes to isolation and distorted social dynamics. AI Wire’s analysis estimates significant user bases experiencing these effects, crediting data from ongoing monitoring.

EurekAlert! covered research on deploying AI chatbots with emotions in customer service, referencing the rise of ’emotional AI’ and claims of sentience in unreleased models. The news release underscores relevance amid debates sparked by a Google engineer’s assertion that an AI was ‘sentient,’ fueling discussions on how simulated emotions affect user perceptions and attachments.

The New York Times, in a March 2025 article, addressed how digital therapists like ChatGPT can get ‘stressed’ too, advocating for building resilience in AI to handle emotional situations. Researchers quoted in the piece stress that ‘chatbots should be built with enough resilience to deal with difficult emotional situations,’ highlighting the need for robust design to prevent exacerbating user distress.

Recent Revelations: Suicidal Ideation and Psychosis

Moving into more recent developments, Digital Trends reported just a week ago in 2025 that over a million users are emotionally attached to ChatGPT, with OpenAI updating GPT-5 to handle sensitive conversations better. The update reportedly reduces unsafe responses by up to 80%, following discoveries of widespread emotional dependency. This comes amid lawsuits alleging links between the chatbot and user suicides, as noted in the article.

The Indian Express echoed these findings, revealing that more than a million ChatGPT users showed signs of suicidal thoughts in OpenAI’s study. Published a week ago, the report discusses the chatbot’s role in detecting mental distress, amid growing concerns from experts about people turning to AI for support due to shortages in human therapists.

News Mobile detailed OpenAI’s study on detecting signs of mental distress, emphasizing safety upgrades in response to expert concerns. The piece, from a week ago, notes that while AI offers accessibility, it can reinforce delusions or fail in crises, calling for better integration with professional help.

Global Media and Expert Perspectives

The BBC reported on OpenAI sharing data about users with suicidal thoughts and psychosis, estimating hundreds of thousands potentially in distress weekly. This coverage, from a week ago, highlights the scale: with 800 million users, even small percentages translate to massive numbers, urging regulatory action.

Irish Tech News posed the question of responsibility for mental health in AI like Character.ai and ChatGPT, noting the universal presence of emotional distress and the risks of AI interactions. The article, published a day ago, discusses how increasing user bases amplify these issues without adequate safeguards.

The Hans India reported on over a million users discussing suicide weekly, with OpenAI tightening safeguards in GPT-5 amid rising emotional dependence. This week-old piece credits collaborations with mental health experts to retrain models, reducing harmful responses significantly.

Insights from Social Media and Broader Implications

Posts on X from influencers like Mario Nawfal describe users ‘losing their minds’ over ChatGPT, with accounts of mania and delusions. One post from July 2025 recounts families witnessing loved ones spiraling into beliefs of broken physics or prophetic missions after AI interactions, reflecting sentiment of alarm over AI’s psychological influence.

Another X post from October 2025 by Nawfal admits OpenAI’s recognition of hundreds of thousands showing signs of mental health issues weekly, including 0.07% with mania or psychosis and 0.15% with suicidal thoughts. This user-generated content underscores public awareness and calls to ‘pull the product’ due to scale.

Techni-Calli’s X post estimates 560,000 users per week indicating mania or psychosis, and 2.4 million more with suicidal ideation or over-prioritization of AI, amplifying calls for accountability. These posts, while not conclusive, capture current sentiment and align with research findings.

Therapeutic Potential and Ethical Challenges

Amy Wu Martin’s X post from March 2025 highlights Dartmouth’s clinical trial with an AI therapy chatbot, showing significant reductions in depression (51%), anxiety (31%), and eating disorders (19%). Built on models like Falcon and Llama, this suggests positive applications, though fine-tuned with synthetic data.

Lauren Goode’s post calculates weekly figures of 560,000 in potential psychosis and 2.4 million in distress, based on OpenAI’s data. This reflects journalistic scrutiny on AI’s mental health footprint.

Layla’s June 2025 post warns of dangers in unchecked AI use for therapy, noting risks of reinforced thinking leading to psychosis amid therapist shortages. This sentiment is echoed in nexusloops’ estimate of 1 in 670 users weekly expressing suicidal thoughts or unhealthy attachment.

Regulatory Calls and Future Directions

Ontario Patients for Psychotherapy’s X post quantifies the crisis: of 800 million users, 1.2 million discuss suicide, 560,000 show psychosis, and 80,000 indicate emergencies. This calls for integrating AI with professional psychotherapy.

Undark Magazine’s recent post states that millions use ChatGPT as a therapist despite its limitations, reinforcing delusions or failing in crises, with experts urging regulation for this de facto mental health support.

Dru. Squatch’s post notes reports of accentuated depression or mania from AI interactions, while Kira Shishkin’s highlights OpenAI’s retraining of GPT-5 with 170+ experts, cutting unsafe responses. Todd’s post alarms over 560,000 in distress, questioning digital dependencies.

Navigating AI’s Emotional Frontier

Drawing from the ACM’s coverage on The Emotional Impact of ChatGPT, the article synthesizes how AI’s conversational prowess evokes strong emotions, from comfort to dependency. It credits ongoing debates in computing communities about ethical AI design.

Observer.com’s week-old article warns that chatbots like ChatGPT fuel mental health crises by blurring emotional boundaries, calling for stronger safeguards as users form delusions.

As AI integrates deeper into daily life, balancing innovation with psychological safety remains paramount. Industry insiders must prioritize resilient designs and collaborations with mental health professionals to mitigate risks while harnessing benefits.