When an AI tool contributes to a clinical error, who is answerable — the doctor, the hospital, or the vendor?

AI has entered Indian clinical practice faster than the rules governing it. Triage tools, radiology assistance, transcription, differential generators, patient-facing chatbots — many of us already use something, formally or informally.

The liability question is unresolved. If an AI-assisted decision harms a patient, the doctor who signed the note is exposed. The hospital that procured the tool may not be. The vendor almost certainly disclaimed responsibility somewhere in its terms.

Three things I would like this assembly to work out:

One — should AI assistance be documented in the record? If a tool contributed to a diagnosis or a dose, does the note say so? There is a reasonable argument both ways, and I don't think our profession has settled it.

Two — what should a hospital be required to disclose before deploying a clinical AI tool: validation data, error rates, the population it was trained on?

Three — what would a workable consent standard look like when AI is involved in a patient's care?

I use AI tools myself and believe they will help us. That is exactly why the profession should define the rules of use before someone else defines them for us, in a courtroom, after a bad outcome.