MACHINES MAY DETECT, OFFICERS MUST DECIDE: ALGORITHMIC ASSISTANCE AND THE LIMITS OF DELEGATION IN GST ADMINISTRATION

India’s Goods and Services Tax (GST) already runs on algorithms: returns are matched by machine, taxpayers are ranked by risk, and invoices are authenticated through a central portal. Within seven weeks in mid-2026, two High Courts quashed GST proceedings tainted by artificial intelligence (AI), and in doing so raised a question that tax law had not previously needed to answer: who is the decision-maker when a machine has done much of the thinking? This article argues that the answer lies in a familiar principle of administrative law rather than in a new body of ‘AI law’. A statutory officer may use an algorithm as evidence or as a drafting aid, but may not allow it to stand in for the application of mind that the statute demands. The article distinguishes predictive analytics from generative drafting tools, shows that the recent decisions address only the second, and warns that a doctrine triggered by visible error will miss competent but unaccountable automation. It proposes an ‘algorithm-removal test’ and a framework of human-accountable algorithmic taxation to give the principle operational content.