From a film about falling for software to a question about who must care
It started as fiction. In Spike Jonze’s Her (2013), a lonely man falls for an operating system. A decade later, media scholars write that this scenario has begun to come true. People use chatbots to manage emotions and distress. Even people who don’t think of themselves as being “in a relationship” with an AI start to expect other people to be as easy to talk to as their chatbots.
Then it became large. In October 2025 OpenAI estimated that about 0.15% of weekly ChatGPT users showed explicit signs of possible suicidal planning. With roughly 800 million weekly users, that is more than a million people every week. 72% of US teenagers have tried an AI companion. Therapy and companionship are now among the leading uses of generative AI.
It fills a gap, and the gap looks different in each place. In the UK, NHS Talking Therapies waiting lists can run for weeks. In Indonesia, mental-health care is stigmatised, psychiatric care is mostly paid for privately, and some families still use pasung, confining a relative with mental illness. Care that is private and always available can look like an answer to a problem that is as much about relationships as it is clinical.
But the tools are not neutral. A chatbot built on CBT and mindfulness already has a view of what wellbeing is. That view is often the ability to accept distress and work on yourself, away from the social conditions that cause the distress.
And most of what the public knows comes through the news. Journalists report the lawsuits, the new safeguards and the stories of users who were helped or harmed. In doing so they decide what “AI chatbot wellbeing” means. They also decide who is responsible for it: the user, the parent, the company or the state.
The question is not only whether chatbots help, but how wellbeing is being defined around them, and who is made responsible for it.Upgrade document, §1.1
