Your organisation is already using AI. The question your board will be asked is not whether you adopted it — it's whether you can account for it.

An agent is granted access so it can be useful, then keeps that access because nothing revokes it. Its reach is usually wider than the task that justified it, and it exercises that reach continuously rather than occasionally.

An AI agent is a non-human identity with initiative.

We establish what each agent can read, write and trigger, whose authority it inherits, and where a single credential spans systems that your controls assume are separate.

Vendors add model-based processing to existing products through release notes. The contract you signed and the processing now taking place may no longer describe the same service, and your data may be leaving a boundary your assessment recorded as closed.

We review supplier documentation, sub-processor positions and integration scopes, and report where AI processing has entered your estate through a third party.

Adoption rarely arrives through a governance forum. It arrives through licences a team already held, browser tooling, and pilots that quietly became dependencies. The result is an inventory nobody maintains.

We build a register of AI in use — sanctioned and otherwise — recording the data categories involved, the owning function, and whether a lawful basis and an accountable owner exist.

Explanation is a model question, contestation a process question. Attribution is neither: it depends on knowing which identity acted, under whose mandate, and on what record.

We test whether that trail exists for the decisions that matter to your regulator and your customers, and identify where it ends before it reaches a person.