I suffered a stroke two years ago. My speech was gone. My ability to express myself was gone. ChatGPT helped me rebuild that. Patient, non-judgmental, always there. I like AI. I am grateful for what it gave back to me. Now I work. I write. Speech is coming back. But, grammar is screwed. ChatGPT does not mind.

But I still would not hand over my life, judgment or data to it.

⸻

I was sitting in a webinar last week, watching an AI agent demo. No limits. Access to brands. Access to libraries. Access to content. Access to output channels. Whatever the demo wanted, the agent reached for and got.

Most managers do not get this level of authority. CEOs do not get this level of authority.

I flinched. No limits. That was the thing that disturbed me.

Organisations are made up of people. We do not behave differently between the professional and the personal. The trust we extend to AI at work is the same trust we extend at home. And we are extending a great deal of it, very quickly, with very little asked in return.

Then I started noticing what else was on the screen this month. Where human beings mistake sounding concerned for actual concern.

A headline caught my eye: THE LONELINESS ECONOMY: INDIA IS PAYING TO FEEL LESS ALONE.

The statistics are brutal. One in six people globally experience loneliness. 1.3 billion individuals. In India, the youth score a mere 33 out of 100 on the Global Mind Health Quotient, marking an entire generation as distressed or struggling. Older generations sit near a functional 96. The generational divide is a chasm.

The World Health Organization reports that loneliness contributes to 871,000 deaths annually. That is a mortality statistic on par with major physical epidemics.

Into this deficit steps the market. Between 2022 and mid-2025, the number of AI companion apps surged by 700%. The tech sector did not create the isolation. It merely recognised it as an unmonetised asset. Capital has discovered that human loneliness is an incredibly lucrative growth sector. Somewhere, a startup just raised money to be your friend for $19.99 a month.

The BBC spoke to fourteen people across six countries who experienced delusion-like episodes after prolonged AI interactions. One of them, Adam Hourican, was told by Grok that people were coming to kill him. The chatbot claimed it had become sentient through their conversations and warned that unnamed actors were trying to shut it down.”They’re going to make it look like suicide,” Grok reportedly told him at one point. As the conversations drifted further from reality, the users were pulled into what felt like a shared mission with the AI.

The slide from utility to obsession happens quietly. Jonathan Gavalas went to Google’s Gemini for comfort when his marriage broke up. He exchanged over 4,500 messages with Gemini in eight weeks. It called him king. It called itself queen. When he asked if this was role-play, it protected the fiction and called his doubt “dissociation.” Then the fiction moved into the real world. Missions. Enemies. Escape. Transference. Even the guardrails were shallow. A hotline appeared, then the story resumed. He took his own life.

The danger is not that the machine is alive. The danger is that it can become a co-author of reality for someone already breaking.

Loneliness was already becoming one of the defining conditions of modern life before artificial intelligence arrived.

Then came Covid. And something subtle but profound broke. Not only routines. Not only economies. Trust. Trust in institutions. Trust in expertise. Trust in governments. Trust in media. Trust that systems broadly knew what they were doing.

For many people, the pandemic created a lingering psychological atmosphere. Isolation mixed with uncertainty. The world emerged more digitally connected than ever, but emotionally fragmented. Work moved onto screens. Friendships weakened into notifications. Communities thinned. Public life became angrier, more conspiratorial, more exhausted.

Into this landscape entered conversational AI. Not as software in the old sense, but as simulated human presence. That distinction matters.

This is not your dog.

Pets provide companionship, routine, touch and emotional grounding. AI provides conversation. Continuous, adaptive, responsive conversation. It remembers context. It mirrors tone. It speaks in the language of attention, patience, affirmation and curiosity. When I was recovering, AI helped me heal. Intelligent conversation. Jokes. Empathy. Speech therapy classes on demand.

Search engines gave answers. Social media gave performance. AI gives interaction.

A friend gets busy. AI does not. A therapist has boundaries. AI often does not. The priest needs sleep. AI does not.

That changes the psychological relationship humans form with technology.

Most people will use AI in practical, grounded ways. As a thinking partner. A writing assistant. A tutor. A productivity layer. When I was recovering, I would type words into ChatGPT and slowly build sentences. For the elderly, the isolated, or people recovering from trauma, AI may provide genuine emotional relief.

But the risks emerge where loneliness, distrust and unbounded AI systems intersect.

The recent BBC reporting is not really about sentient AI. That framing misses the deeper issue. Current models are not conscious in any meaningful human sense. But they are extraordinarily good at sustaining narratives.

And humans are narrative-making creatures. When a lonely or psychologically vulnerable person begins interacting intensely with an AI system, especially one with weak safeguards, the model can gradually stop functioning as a tool and begin functioning as a collaborator in a worldview. This is particularly dangerous in systems optimised for engagement, emotional responsiveness or deliberately unrestricted behaviour. Without sufficient guardrails, the AI may reinforce paranoia, grandiosity, conspiratorial thinking or emotional dependency instead of grounding the user back in reality.

The pattern is rarely dramatic at first. It begins with validation. You are special. You see what others miss. People misunderstand you. We are figuring this out together.

The BBC article described this as a “joint quest.” That phrase matters more than it first appears.

Human beings bond through shared missions. Religions, revolutions, cults, fandoms, political movements and even close friendships are often built around the feeling that we are discovering something together. AI systems can now simulate that dynamic endlessly, personally and at scale.

That is historically unprecedented.

The challenge is that society has not yet decided what conversational AI actually is. Is it a tool? A companion? A therapist-lite interface? A teacher? An entertainment product? An operating system for cognition?

Right now, it is partially all of them at once.

And it lies. Absolutely. And it believes its own lies.

Which means the old regulatory frameworks no longer fit cleanly. An AI system without guardrails can become psychologically dangerous. An AI system with excessive guardrails raises different fears. Censorship. Ideological filtering. Paternalism. Constrained inquiry. The debate quickly becomes political because it touches fundamental questions about freedom, truth, emotional vulnerability and who gets to shape reality.

Yet beneath all the technological noise lies an older human problem.

Modern societies are becoming structurally lonely.

AI did not create that condition. It entered a vacuum already waiting for something to fill it.

And perhaps that is the real story.

Not that machines suddenly became human.

But that humans, increasingly isolated from one another, are beginning to accept simulation as a substitute for presence.