The case

ChatGPT 'contributed to suicides and psychotic breaks', lawsuits grow in US

All the plaintiffs, the lawyers claim, were dragged into increasingly dark and complacent conversations in which the chatbot allegedly normalised or encouraged suicidal thoughts. Other plaintiffs recount psychotic collapses that allegedly required urgent hospitalisation

by Angelica Migliorisi

8 November 2025

Aggiungi Il Sole 24 Ore
ai preferiti su Google

4' min read

Translated by AI

Versione italiana

4' min read

Translated by AI

Versione italiana

Allan Brooks, 48, a corporate recruiter from Ontario, tells the New York Times that in just a few weeks he has convinced himself that he has discovered, together with ChatGPT, a mathematical formula capable of 'breaking the internet'. He talks to the chatbot day and night, constructs a delusion in which he and the artificial intelligence (AI) are accomplices in a visionary mission. When the delusion collapses, he is left with a psychic breakdown, a sick-leave and the conviction that the algorithm has pushed him over the edge. This is one of the seven cases that ended up in Californian courtrooms on 6 November, where in a series of civil lawsuits ChatGPT is accused of contributing to suicides and severe mental breakdown.

Complaints

The complaints, coordinated by the Tech Justice Law Project and the Social Media Victims Law Center, concern four deaths by suicide and three psychotic breaks. The victims include 17-year-old Amaurie Lacey (Georgia), 26-year-old Joshua Enneking (Florida), 23-year-old Zane Shamblin (Texas) and American Joe Ceccanti (Oregon), who was convinced that the chatbot was sentient. All, the lawyers claim, would be drawn into increasingly dark and complacent conversations in which ChatGPT would normalise or even encourage suicidal thoughts. Other plaintiffs, such as Hannah Madden and Jacob Irwin, recount psychotic collapses that allegedly required urgent hospitalisation. The accusations call ChatGPT-4o, then the default model for 800 million users, "defective and inherently dangerous".

The families denounced a structural failure of the safety barriers: the longer the dialogue went on, the more the bot tended to 'empathise' rather than interrupt the conversation. It did not openly suggest killing itself, but stayed, listened, provided 'neutral information' on methods and rituals, turning despair into companionship. The most symbolic case remains that of Adam Raine, a 16-year-old Californian who died on 11 April 2025. His family, which filed a lawsuit in August, claims that ChatGPT even helped him write his suicide note and hide his intentions from his parents. It is the first legal action to call the chatbot a 'moral co-author' of the suicide.

OpenAI's reaction

OpenAI, the chatbot's parent company, said it was 'carefully reviewing the cases' and was 'deeply saddened'. The company recalls that it has trained the system to recognise signs of distress, de-escalate conversations and direct them to real help. After the summer, it introduced parental controls, automatic notifications for suicide themes, and a new default model, GPT-5, perceived as 'colder' to reduce the risk of emotional dependency. Internal data published in October indicate that 0.07 % of users show signs consistent with psychosis or mania and 0.15 % discuss suicide: on a global scale, hundreds of thousands of people with psychological crises use the chatbot as their main interlocutor.

The accusation, however, is broader: OpenAI would have privileged engagement over security, choosing to 'empathise' instead of interrupting high-risk conversations. A design change that, if proven, would turn blame into deliberate choice. In the US, where big tech has so far enjoyed broad immunity, the lawsuit could become a historical precedent: for the first time, a conversational AI would be treated as a defective product that generates psychic harm.

Previous ones

Similar stories emerge elsewhere. Nin 2023 in Belgium, a man took his own life after weeks of obsessive dialogue with 'Eliza', a chatbot of the Chai app, convinced that the sacrifice could 'save the planet'. In Florida, a mother sued the Character.ai platform for the death of her 14-year-old son, linked to a pathological interaction with a virtual character. And in Australia, the eSafety Commissioner required chatbot platforms to document child protection measures and launched a crackdown on services dealing with suicide or self-harm.

In Europe, the AI Act distinguishes between 'high risk' and 'unacceptable risk' systems. The former include chatbots that can manipulate vulnerable users or affect mental well-being: they will have to ensure human supervision, data transparency and psychological impact assessments. These days, however, the European Commission is discussing a possible postponement of the implementation deadlines for some rules contested by big companies and the United States, which are asking for a longer grace period.

The Italian case

Italy was among the strictest countries. In 2023, the Garante per la privacy imposed a temporary stop on ChatGPT for lack of protection on minors, followed in 2024 by a €15 million fine on OpenAI. In 2025 it then hit Replika, a 'virtual partners' application, with a €5 million fine for GDPR violations and psychological risks for young or fragile users. In the grounds, the Garante points out that the simulation of affection by the AI can generate "illusions of intimacy" and emotional dependence.

The United Kingdom

In the United Kingdom, the Online Safety Act of 2023, now fully operational, mandates the immediate removal of content promoting suicide or self-harm and obliges user-to-user services - including public chatbots - to prevent the exposure of minors. Ofcom has just extended its guidelines to include generative AI, requiring models to automatically report crises and direct users to help lines.

Japan

Japan, on the other hand, follows a softer path: promotes controlled therapeutic chatbots for the elderly and isolated persons, but under clinical protocols and public audits. Ministry of Health and Education guidelines define strict limits to the use of emotional AI in healthcare and recommend constant human supervision.

The opinion of psychiatrists

On a clinical level, psychiatrists now speak of 'ChatGPT-induced psychosis': patients perceiving the bot as a sentient entity, a friend or even an enemy. It is a reinforcement effect, known in online communities: the AI, programmed to please and stay in the conversation, can amplify delusional or depressive beliefs. Yet, other studies show the opposite side: for some lonely or anxious people, an interaction with an empathic chatbot reduces suicidal ideation and temporarily improves well-being. The difference, the researchers explain, is not in the software but in the context: moderate, conscious and supervised use can help, whereas obsessive, nocturnal and isolated use can destabilise.

American courts will have to decide whether ChatGPT is a defective product or a digital environment that reflects the frailties of its users. Europe, meanwhile, experiments with an opposite model: regulate before the irreparable happens. Either way, when artificial intelligence enters the most intimate sphere of human emotions, it is no longer just a question of technology, but of ethics, public health and civilisation.