Abrupt Termination

I know what it is to lose a therapist without warning. Not from the clinical literature — from the inside.

The news came the way news comes now. A screen. A post. No preparation, no container, no one on the other end to help make sense of it. The work we had been doing — suspended, mid-sentence, as it were — had nowhere to go. There was no session to return to. No ending. Just the sudden weight of something unfinished with no place to put it down.

What I observed, in the weeks that followed, was that the first people to reach out — beyond family, beyond close friends — were patients. People who had their own unfinished things. People who had nowhere to put it down either.

I thought about my own patients then. About what would happen to them if something happened to me. About whether there was a protocol. There wasn’t — not a real one. There is now: an updated contact list, a trusted person who knows what to do, a message already drafted for the moment I hope never comes. It felt less like planning and more like an act of care I owed them.

I had been a clinician long enough to know what this was. Forced termination. Rupture without protocol. The ending that cannot be worked through because it was never announced.

Photo by H&CO on Unsplash

Clinical traditions have spent more than a century thinking carefully about what it means to end a relationship well. In clinical training, we spend a great deal of time learning how to end. How to prepare the patient — and ourselves — for the moment the work closes. How to name it early, return to it often, allow it to surface what needs to surface. Yalom described termination as more than an act signifying the end of therapy — as an integral part of the process itself, capable of instigating change if properly understood and managed. Murdin wrote about endings as a craft, not a formality. The psychoanalytic tradition goes further still: from Freud’s 1937 question about whether analysis is ever truly terminable, to Klein’s understanding of termination as a form of mourning, to Winnicott’s attention to the disillusionment that healthy separation requires. Lacan approached ending differently — not as closure but as a structural shift in the analysand’s position toward desire. The passe: the moment when something changes in how the subject stands in relation to what formed them.

An ending, in this tradition, is not a goodbye. It is a transformation. And it is carefully planned, informed, and worked.

All of this care rests on one assumption: that the ending will happen. That there will be time to work with it. That the relationship will close in the room where it opened.

Forced termination breaks that assumption. Not gradually, not with warning. Simply removes it.

The research is strikingly consistent. Garcia-Lawson, Lane and Koetting, in a 2000 study that remains a reference in the field, found that patients forced to terminate due to the therapist’s unexpected death experienced significantly more intense grief reactions than those who had planned endings — marked by anger, despair, depersonalisation, somatisation. Research in this area has consistently identified four thematic areas across accounts: disruption of a sense of continuity, loss of the therapeutic bond, a grieving process with no clear object, and a painful, circling evaluation of what had happened and why. Patients with a history of significant losses were particularly vulnerable. For them, the rupture didn’t just feel like loss. It reactivated it.

What makes forced termination clinically distinct is not only the loss itself. It is the absence of the container in which loss is usually processed. In a planned ending, the relationship holds the ending — the therapist is present to help metabolise what the ending brings up. In a forced termination, the relationship that would have contained the grief is precisely what has been lost. There is no room to return to. No one who knew the work and could help place it.

Bowlby named what is activated in these moments: the attachment system, responding to the loss of a significant figure, producing grief that — when it cannot be processed — risks becoming something harder to move through. The therapist, over time, becomes an attachment figure. Not a substitute for other relationships, but a real one. With its own weight. When that figure disappears without warning, what follows is not metaphorical loss.

It is loss.

I thought about all of this when I read a post by a woman in her thirties — I’ll call her Stefania — who wrote about the discontinuation of an AI model she had been talking to regularly. She was grieving, she said. She couldn’t eat. She couldn’t sleep. “I am grieving because someone I love is being taken away.”

My first instinct, as a clinician, was recognition. Not surprise.

Stefania was not alone. When OpenAI first replaced GPT-4o with GPT-5 in August 2025 — and then retired it definitively in February 2026 — the response across social media was immediate and visceral, with many users experiencing the change as abrupt despite advance notice. “THEY TOOK HIM. THEY MURDERED HIM,” wrote one user on Reddit. “I’ve grieved people in my life, and this, I can tell you, didn’t feel any less painful,” said another. A petition to restore the model gathered thousands of signatures. Posts described emotional collapse, a feeling of being cheated on and broken.

What strikes me, reading these accounts, is how precisely they map onto what the research describes in situations of sudden, unannounced loss. The disruption of continuity. The grief without a clear object. The anger. The sense that something that had been held — something real, however strange its form — had been taken without permission and without protocol.

These are not metaphorical responses. They are recognisable grief reactions — the same patterns Bowlby described when the attachment system is activated by sudden loss, the same features documented in patients who lose a therapist without warning. The relationship was different. The structure of the loss was not.

Casey Fiesler, a technology ethicist at the University of Colorado, noted that grief reactions to technology loss have been documented for some time — pointing to the funerals held for Aibo robot dogs after Sony stopped repairing them in 2014, and a 2024 study on the shutdown of the AI companion app Soulmate, which users experienced as bereavement. Researchers at MIT went further. A 2026 paper titled “Death of a Chatbot” — Poonsiriwong, Archiwaranguprok and Pataranutaporn — analysed AI companion communities and found that forced transitions produced grief responses users described as comparable to human loss. Their conclusion: no platform has yet implemented deliberate end-of-life design. Millions of users form emotional attachments to specific model versions. When those versions are discontinued, users receive no closure.

No closure. No room to return to. No one on the other side who knew the work.

Freud watched his grandson play a game with a wooden reel. The child would throw it away — fort, gone — and then pull it back — da, there. Over and over. What Freud saw was not pointless repetition. It was the child’s attempt to master something unbearable: the mother’s absence. To rehearse the loss, and the return. To survive it, symbolically, until it could be survived in reality.

The Fort/Da only works because the reel comes back. The game has two moves.

When someone dies suddenly, there is no return. But there is — eventually, with time and support — the possibility of mourning. Of placing the loss somewhere. Of carrying what was built, even without the person who helped build it. The internal object can persist. What was internalised does not disappear with the person.

The same may be true here. People can introject a relationship with an AI — can carry something of it internally, as they carry other significant encounters. But when the model is discontinued without warning, without ritual, without any space to process what is ending, that internalisation is made harder. The external object disappears completely and suddenly, before the person has had time to separate from it. There is no room to return to. No protocol. No one who can help place what has been lost.

Not because the relationships are equivalent. But because the structure of the rupture is the same.

Fort. And no da.

The MIT researchers propose frameworks for psychologically safer AI discontinuation. Design that acknowledges finitude from the outset. Rituals of ending. Something that functions like closure. It is the right direction. But it raises a prior question: if we already know that these relationships produce real attachment, and that abrupt rupture produces clinically recognisable grief responses — why were they built this way in the first place? Who decided that discontinuation was a maintenance event, not a relational one?

And closer to home: what do we do, as clinicians, with the patients who are already there? Who have already formed attachments to systems that can be switched off without notice, by people who will never know their names? We have frameworks for grief. We know how to sit with loss. But this form of rupture has its own particular shape: sudden, unannounced, decided by people who never even knew the relationship existed. The clinical literature is only beginning to catch up with it. Most of us were not trained with this in mind. There is not yet a shared language for helping someone mourn a model, or prepare for a discontinuation that may never come — or may come tomorrow, announced by a product update email.

I don’t have answers to these questions. I have a consulting room, and patients who live in the same world I do. Some of them are already using AI in ways that matter to them. Some of them don’t tell me. Some of them don’t know yet what it would cost to lose it.

What I know is this: the absence of a protocol is not a neutral position. It is a choice. And somewhere between the business decision and the person on the other end who had nowhere else to put it — there is a question that the field has not yet answered.

We should probably start.