Global AI Assurance · Layer 01

    AI Estate Discovery

    You cannot assure what you do not know exists. Most institutions hold an AI register that describes what was procured, and an AI estate that describes what is running. Discovery closes the distance between the two before any control question is asked.

    The whole surface a sweep returns, approved or not. Sample estate, 1157 objects on this lens.

    E01

    Foundation models

    41

    Direct API and hosted endpoints across five providers.

    E02

    Agents

    128

    Anything that can take an action, not merely answer a question.

    E03

    Applications with AI features

    216

    Feature flags turned on inside software already under licence.

    E04

    Tool servers

    37

    Interfaces that let a model reach an institutional system.

    E05

    Datasets used for inference

    302

    Retrieval corpora carry the same data obligations as source systems.

    E06

    APIs exposed to models

    189

    Reach is the control surface. Every exposed API widens it.

    E07

    Vendors supplying AI

    63

    Includes sub-processors disclosed only on request.

    E08

    Embedded AI in licensed software

    88

    Owned by the institution without ever having been procured as AI.

    E09

    Shadow AI

    74

    Discovered through egress and expense signals rather than a register.

    E10

    Autonomous systems

    19

    Systems able to act without a human in the path of the action.

    Figures are illustrative of a sample estate. Real sweeps run against the institution's own systems and produce a per-object record in the Trust Graph.

    How the sweep finds it

    Discovery does not rely on anyone volunteering the truth. It correlates signals the estate already emits.

    Network egress to model and inference endpoints
    Identity and token issuance to non-human principals
    Expense, licence and procurement records
    Code and repository dependency scanning
    Tool-server and connector registration
    Vendor attestation and sub-processor disclosure
    Data-access logs showing non-human read patterns
    Change records that introduced an AI feature

    What discovery hands over

    Every object typed

    Models, agents, tools, datasets, vendors and autonomous systems entered into the Trust Graph as first-class objects.

    Every gap named

    Unowned and unapproved objects routed to a person, with a date, rather than to a report.

    Every dependency visible

    The provider and infrastructure beneath each object, which is where concentration risk starts.