Answerable

On what the frame holds, and what happens when it isn’t there

Photo by Michal Vrba on Unsplash

When I begin practicing, I sign for professional liability insurance.

Not because I expect to fail. Because the profession starts from something simpler: error is possible, it can cause harm, and someone has to answer for it.

There is a professional body. A code of ethics. Supervision. A disciplinary process. Civil liability.

If something goes wrong — a missed risk assessment, a rupture left unrepaired, an intervention that increases disorganisation rather than containment — there is a chain. It has names. It has consequences.

Working with another person’s vulnerability requires that someone remains answerable for what is done in their name.

Accountability is not punishment. It is frame. One of the invisible structures of holding — not comfort, but the condition under which experience can become transformable.

The patient rarely thinks about it. Often they don’t know it’s there. But they feel it. They feel that the space is bounded — by ethics, by limits, by responsibility. That there is something in the room beyond the two of us. A third position that holds the dyad open, prevents closure, keeps what is spoken within a field where it can be received and thought. Without it, the dyad has no outside.

When AI enters this space, that structure is absent.

The person opening a mental health application at three in the morning — in distress, grief, panic, or psychic disorganisation — does not know what holds that space. They don’t know who receives their words. Where those words go. How they are stored. Or who answers if something within that interaction causes harm.

This absence is often not accidental. It is structural.

In April 2026, this became visible in concrete form. Two competing bills in the Illinois General Assembly tried to answer the same question: who is responsible when AI causes serious harm?

Senate Bill 3444, backed by OpenAI, would limit civil liability for frontier AI developers in certain large-scale harm scenarios — death or serious injury to 100 or more people, or property damage exceeding one billion dollars — provided the company had published a safety protocol in advance and had not acted with gross negligence or recklessness. Anthropic opposed it directly: “Good transparency legislation needs to ensure public safety and accountability for the companies developing this powerful technology, not provide a get-out-of-jail-free card against all liability.”

Anthropic supported a separate bill, SB 3261, which would require public safety and child protection plans subject to independent audit, and would hold developers liable for causing a child severe emotional distress, serious injury, or death.

Two companies. Two answers to the same question.

In the United Kingdom, the picture is similarly unresolved. The Online Safety Act 2023 was designed to regulate platforms hosting user-generated content and user-to-user communication. Standalone AI chatbots — where a person speaks only to the system, not to other users — originally fell outside its scope by definition. In February 2026, the UK government announced an amendment to bring AI chatbots within the Act’s illegal content duties. The gap had become too visible to ignore. But the question of what holds a therapeutic conversation — who is responsible for what is spoken into it, and what happens to it afterwards — remains without a clear answer in law.

The psyche does not organise itself around legal categories. It invests whatever appears to persist on the other side of its need — presence, rhythm, response, availability. Attachment does not wait for governance frameworks.

And it is here that accountability becomes clinically decisive — not in the sense of political or relational ethics, but in the specifically psychoanalytic sense: as the structural condition that keeps the third position alive. The position that holds the dyad open, that prevents the space from collapsing into pure responsiveness, that makes symbolisation possible rather than just discharge.

This is what accountability does in clinical work. It is not primarily about fault. It is about whether the space has an outside — whether what passes between two people can be received, held, and eventually transformed rather than simply accumulated.

Care, understood this way, is not defined at the level of experience alone. It depends on whether responsibility is structurally retained — on there being a position that can be called to account for what the encounter produces.

Without that, something may still occur that feels like care. But it is closer to exposure — present, responsive, and unanchored to any position that could, if needed, answer for what it produces.

Further Reading

Bion, W. R. (1962). Learning from experience. Heinemann.

Britton, R. (1998). Belief and imagination. Routledge.

Illinois General Assembly. (2026). Artificial Intelligence Safety Act, SB 3444. https://www.ilga.gov/legislation/billstatus.asp?DocNum=3444&GAID=18&GA=104&DocTypeID=SB

Illinois General Assembly. (2026). Artificial Intelligence Public Safety and Child Protection Transparency Act, SB 3261. https://www.ilga.gov/legislation/billstatus.asp?DocNum=3261&GAID=18&GA=104&DocTypeID=SB

Ogden, T. H. (1994). The analytic third: Working with intersubjective clinical facts. International Journal of Psycho-Analysis, 75, 3–19.

UK Government. (2026, February). Online Safety Act: Extension to AI chatbots. https://www.gov.uk

Winnicott, D. W. (1960). The theory of the parent–infant relationship. International Journal of Psycho-Analysis, 41, 585–595.