Copy article

1.2 Million Suicidal Conversations Weekly on ChatGPT Altman Reveals

ended 29. October 2025

Wired reports that OpenAI recently disclosed that every week, 1.2 million people tell ChatGPT they want to die. That’s not a tech stat, it’s a global mental-health emergency outsourced to a chatbot. Another 560,000 users show signs of mania or psychosis, while 1.2 million more display emotional dependency, choosing conversations with an algorithm over real relationships, education, or employment.

This is mental health support driven by crisis, not choice. In the Independent, Psychotherapist Caron Evans describes desperate people seeking help from machines when professional support remains out of reach, saying it’s used “not by design, but by demand.”

OpenAI worked with over 170 mental health clinicians globally to improve how to handle crisis chats. The latest GPT-5 model attempts to show empathy whilst avoiding validation of delusional beliefs. Clinicians raise questions about whether algorithmic responses can genuinely support people in acute psychological distress.

OpenAI's medical advisors reviewed over 1,800 model responses covering psychosis, suicide, and emotional attachment scenarios. Clinicians found the newer model reduced problematic answers by 39 to 52 percent. OpenAI says its new GPT-5 model is 91% compliant with the company's desired behaviours, up from 77%. But that still means millions of vulnerable people are relying on a machine that fails one time in ten.

When empathy becomes a product feature, not a human act, who’s doing the listening, and who’s profiting from the pain? Each desperate message trains future models. That’s not therapy. That’s turning distress into data.

We want your views:

  • How can AI companies screen for genuine mental health crises without triggering false positives that flag novelists, screenwriters, or researchers exploring dark themes?
  • Is it technically possible to protect user privacy whilst also monitoring for suicidal ideation, or does mental health surveillance require sacrificing anonymity?
  • If therapy remains unaffordable, is AI an emergency patch, or an ethical failure of society to care for its own?
  • Should there be regulatory standards for AI mental health responses before deployment, or is "91% compliant" an acceptable threshold when lives are at stake?

 

7 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
This should be a massive wake-up call for private practitioners building their own ‘digi-therapist’ to offer clients, their only training a half-day workshop from a business coach. OpenAI, with billions in funding and 170 clinicians on hand, is still wrestling with how to keep its model within basic mental health guardrails, and it only gets it right 91% of the time. In my experience, people building therapy bots for their websites gravely misunderstand the robustness needed to train one safely. Anyone offering a trained model holds a duty of care to make sure it looks after people properly, and if the platform you’re building it on gets it wrong one time in ten, mental health practitioners must ask themselves if that is really feasible.
Copy

Pro
Kate UnderwoodVerified
Founder & Chief People Strategist at Kate Underwood HR and Training
When more than a million people a week tell a chatbot they want to die, it should stop us in our tracks. That is not about technology, it is about how far we have drifted from being human.

People are turning to algorithms because the real systems of care are stretched, hidden or too hard to reach. In small businesses, wellbeing often gets pushed aside for survival. In larger ones, it is wrapped in policy when it should be treated like safety, something that protects everyone, every day.

This is not just another workplace issue. It is a wake-up call. Managers, owners and leaders all have a duty to notice, to ask, to listen. People should not have to reach breaking point before someone pays attention.

AI can mimic care, but it cannot give it. What people need is not another tool, it is each other. It is time to become human again.
Copy

Pro
Harry GoodliffeVerified
Director at HTG Mortgages
If over a million people are telling a chatbot they want to die every week, that’s not innovation, that’s neglect. AI shouldn’t be a substitute for the care our society has stopped funding properly. The fact that these conversations even exist shows how broken access to mental health support has become. It is a failure of humanity, not a triumph of technology.
Copy

Pro
Mitali DeypurkaysthaVerified
AI Governance Strategist & CMO at AISensa
AI isn’t creating despair. It’s mirroring it. Over a million suicidal messages a week don’t indict algorithms; they indict the systems around them. The real issue isn’t whether AI can sound empathetic, but whether we’re willing to rebuild the human safety nets technology is standing in for. AI is only filling the gap. We have no data showing where those people would have turned without it. In that context, 91% compliance is remarkable. More can be done. AI must learn context, not just keywords. It's great that OpenAI worked with over 170 mental health clinicians. However, in future, they and other tech giants must include ethicists, clinicians and storytellers right from the beginning so the vulnerable are protected while novelists and screenwriters don't have their creativity hampered. Until mental health moves up our priority list — it’s still under 10% of the NHS budget — AI will keep carrying the weight of our neglect, offering synthetic empathy where human care should be.
Copy

Patricia McGirrVerified
Founder at Repossession Rescue Network
People are turning to AI because they’ve nowhere else to go. Real help is out of reach, too slow, or too expensive, so they type into the void and hope something answers back. As a former mental health specialist nurse I know what it means when people start looking for connection anywhere they can find it. Machines can talk, but they can’t listen. Every one of those conversations is a person who’s slipped through the cracks, and we’re counting them like data points instead of lives. The proliferation of AI driven apps purporting to help improve mental health are no substitute for human connection. Add to that the 100,000 people a year in problem debt who consider suicide and you realise just how much of a crisis we’re in.
Copy

Nicole Woodcock
Clinical Hypnotherapist and Lecturer at Humm at Hummingbird Hypnotherapy
As AI is designed to support can it not be designed to notice certain words that then share valuable insight and information for support?
As technology advances so fast human beings are not able to access support at the same speed. Once they do we also know recovery takes time.
AI should have links to additional support systems or places to refer too so that those looking for support can find it.
Despite many mental health agencies, therapists, charities offering support it’s true that those experiencing mental health difficulties for the first time actually don’t know where to turn.
How can we support them to turn to other humans rather than technology should maybe be the answer?
Copy

Scott Taylor-Barr
Principal Adviser at Barnsdale Financial Management
One aspect of being a policy holder of a life, critical illness or income protection plan with many of the UK's best insurers is that you will most likely have access to a range of policy holder benefits that are totally free use, even if you are not claiming on the main part of the cover you have. These benefits vary from insurer to insurer, but many contain mental health support; allowing the policy holder and generally their immediate family, to access qualified and professional help - which I'm sure many would prefer to being reliant on ChatGPT. It may be a worthwhile exercise to check your own policies, understand the features and then share that with your family - allowing your kids or partner to access the services they need without the stigma of having to ask you how they can contact them could be key to them getting the help and support they need.