Answerability
The criterion on which an AI deployment in professional or adjudicatory work is accepted or refused: whether, after the machine has acted, an accountable person remains who can be held to answer for the work. A machine may verify, but a machine cannot be a respondent; responsibility cannot be delegated to what cannot be held to account.
When this decides something for you
A court administration is offered a system that drafts orders. A regulator is asked whether a firm may run a service in which the machine does the research. A Bar Council is asked whether a tool that proposes authorities may be used in filings. Each body needs one test that separates deployments it should accept from deployments it should refuse, and "is the machine accurate?" is not that test, because accuracy is a property of the tool and the profession's duties attach to people. Answerability is the test.
The question
On what criterion does a body with the power to refuse an AI deployment decide to refuse it?
The premise
Every profession exists because someone must answer for the work: to a court, a regulator, a client, a patient. A machine can produce the work. It cannot appear, be examined, be disciplined, be struck off or be sued to any purpose. Whatever it does, the answering remains with a person. A deployment that removes the person who answers has removed the profession, whatever it has done to the output.
How it works
The criterion asks one question of any proposed deployment: once the machine has acted, is there still an accountable person who framed the task, verified the output and owns the result, and can be held to it? If yes, the deployment is within the profession's answerability and the body has no ground to refuse it on this criterion, whatever else it may require. If no, the deployment is refused, and the refusal is principled rather than protective: it does not defend the practitioner's income or the old way of working; it defends the one feature of professional work that the machine cannot supply.
Three consequences follow. First, the criterion is indifferent to accuracy. A highly accurate system used in a way that leaves no one answerable fails; a mediocre system used under a supervising professional passes, because the professional answers for the mediocrity. Second, the criterion is what makes verification a duty rather than a preference: a person who has not verified cannot answer, so a deployment that makes verification impossible fails. Third, the criterion locates the refusal power correctly. The body that can refuse is the body to which the professional answers, because it is the body that would be left without a respondent.
The criterion is drawn from the asymmetry that underlies the rest of the method. A machine may check a citation against a database; that is verification of a kind. A machine cannot stand before a disciplinary committee and explain why it relied on what it relied on. The first is useful. The second is what the profession is for.
One case
Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd, 2026 INSC 668, Supreme Court of India, 2 July 2026. The Court, having found that the National Company Law Tribunal relied on precedents that were non-existent or misattributed and that the respondent's own counsel had not cited them, declared zero tolerance for such reliance at the Bar and on the Bench, then said at paragraph 8: "We are aware that mere declaration of prohibitory action is not sufficient; there must be a consequential action following accountability." It directed the Bar Council of India at paragraph 9 to prescribe a guiding principle "along with the disciplinary action that will follow a violation of the norms". That is the criterion applied by a court: a prohibition without a person who answers for its breach is a declaration, not a rule, and the body that holds the professional to account is the one directed to write it.
The rule
Accept a deployment that leaves an accountable person who framed, verified and owns the work. Refuse one that does not. Accuracy is not the test; answerability is.
Not to be confused with
Answerability is not liability allocation between vendor and user, which contracts do, and not explainability, which concerns whether a system's output can be understood. It is the prior question: whether, whatever the output and whoever pays for it, a person remains who can be called to answer.
Provenance and method.
Where this was published
- Shivam Shukla
- The Power to Say No: Activating the Missing Accountability Layer of the Solidarity Stack
- SSRN 7454698 (2026)
- sections 4 and 8
Changelog
- 2026-09-25, version 1. First published.