Case: the AI system nobody had listed
A mid-sized insurer began its ISO 42001 journey confident it used AI in two places: a chatbot and a pricing model. The inventory exercise found nine.
The other seven had crept in unnoticed. A marketing team using a generative tool on customer data. A spreadsheet macro quietly upgraded to a machine-learning plugin. A recruitment filter embedded in an HR platform, screening candidates by criteria no one had reviewed. Each adopted locally, for good reasons, with no governance and no one accountable.
ISO 42001 starts, like most management systems, with knowing what you have. You cannot govern, assess or oversee AI systems you don't know exist — and shadow AI proliferates precisely because these tools are easy to switch on and invisible once running.
The recruitment filter was the wake-up call: it turned out to be quietly disadvantaging a protected group, a legal and ethical exposure no one had chosen. Building the inventory wasn't bureaucracy. It was the moment the organisation could finally see, and govern, what it was already doing.