ISO 55001 and AI agents: how far can we hand over control?
Abstract
Artificial intelligence agents are no longer limited to producing analyses or answering questions. In some industrial facilities, they examine alarms, search for the causes of a failure, prepare interventions, create work orders and can even act directly on the process. This development promises faster decisions, better-targeted maintenance and more efficient operations. But it raises a question that performance gains cannot resolve: when a decision concerning an asset has been prepared or executed by a machine, who can still explain it and who must answer for it? The issue is not whether artificial intelligence is competent in the technical sense. It is whether the organisation remains able to demonstrate how the decision was made, from what data, within what limits and under whose authority. This is precisely where asset management meets the governance of artificial intelligence. ISO 55001 does not mention AI agents, but it requires organisations to control their decisions, their information, their responsibilities, their suppliers and their risks. The technology changes; these obligations remain. Author preprint, English version translated by the author. French original: 10.5281/zenodo.23195792; Arabic version: 10.5281/zenodo.23195796.
