The Answerable Systems Framework: Vedāntic Guidelines for Accountability in Artificial Intelligence

Responsible-AI governance has converged on transparency and explainability as its central instruments. This paper argues that the convergence is incomplete in a specific way: a system can be fully explainable and still leave nobody obligated to respond when it causes harm. That is the answerability gap. A system is answerable when information adequate to evaluate a consequential effect reaches a named party who holds the authority, competence, time, and duty to respond — including the duty to correct, restrict, or withdraw the system. The paper defines answerability as the conjunction of four conditions — informational sufficiency, located obligation, effective authority, and enforceable consequence — deriving the structure from Schedler (1999) and Bovens (2007) and extending it from actors to sociotechnical arrangements. It then develops the Answerable Systems Framework, whose working core is the Eight Tests of Answerability (purpose, authority, evidence, affected persons, limits and traceability, contestability, institutional memory, and repair), as a non-compensatory assurance instrument in which a red line on any test blocks consequential deployment. A third contribution draws four constructs from Vedāntic practical philosophy — viveka, anāsakti, dharma, and loka-saṃgraha — as instruments for the institutional failure that documentation does not prevent: attachment to one's own system. The framework is applied to agentic AI and mapped against NIST, EU, UNESCO, and OECD instruments. Limitations and falsification conditions are stated.