AI Agents Must Not Become Undeclared Office-Holders

An institution authorises an artificial-intelligence agent to prepare decisions, retrieve records, communicate with users and carry out approved tasks. The system is described as an assistant. Yet it can cross databases, choose a route, trigger another tool, alter what people see and act before a leader reviews every step. Its practical authority may exceed that of many employees whose roles appear on the organisation chart.

The language of assistance can hide a governance change. A person who exercises institutional power normally holds an office, mandate, supervisor, set of permissions, professional duty and route of accountability. An AI agent can acquire parts of that power through software configuration without receiving any of the constitutional disciplines attached to office.

AI agents must not become undeclared office-holders. Values-based leadership should treat their actions as exercises of delegated institutional authority. Purpose, scope, permissions, approval, evidence, intervention and remedy must remain visible even when the work is fast, distributed and technically complex.

At 7:02 East Africa Time on 15 October, Reuters reported that the Financial Stability Board had called for greater human oversight of agentic AI in final guidance for banks and other financial institutions. Reuters says the final version strengthens the treatment of agents, calls for more intensive oversight in higher-risk uses and recommends defined triggers for obtaining human approval. It also removes draft language that described AI agents…