Playbook
Stand Up an AI Bill of Materials
Why this is hard
An AI system is a supply chain: models, agents, the tools they can call, the identities acting, and upstream dependencies. Most organizations have no single inventory of it, which means they cannot bound what they run, trace a compromised component, or attest to an AI system for an auditor. Building an AI-BOM turns a vague sense of exposure into a concrete, reviewable list.
- Agents and MCP servers proliferate faster than any manual list
- A compromised dependency is only traceable if it was inventoried
- Attesting to an AI system requires enumerating its parts
- The gap between the inventory and observed actors is your shadow AI
The approach, step by step
From no inventory to a living AI-BOM
Enumerate the models
List the models and model integrations in use across the platform, including which systems and agents call them.
Catalog the agents and tools
Record each agent, what it is scoped to do, and the MCP servers and tools it can invoke.
Map non-human identities
Inventory the service accounts and tokens acting as non-human identities, and note the human owner of each.
Trace the dependencies
Capture the upstream dependencies in the AI supply chain so a compromised component can be located later.
Reconcile with reality
Compare the inventory against the actors actually observed acting, so the gap, the shadow AI, is explicit and closed.
How Fintra Trust does the work
Fintra Trust treats the AI-BOM as living, not a one-time spreadsheet. Because it governs actions, observed non-human actors are reconciled against the inventory automatically, so an agent or MCP server that is acting but not catalogued surfaces as shadow AI. The AI-BOM then feeds control mapping and vendor reviews.
What you get out of the box
- An inventory of models, agents, MCP servers, and dependencies
- Non-human identities tied to their human owners
- Automatic reconciliation of observed actors against the BOM
- A foundation for control mapping and vendor security reviews
Avoid the common pitfall
Questions people ask
What is an AI-BOM?
An AI Bill of Materials is a structured inventory of the components an AI system depends on: models, agents, MCP servers and tools, non-human identities, and upstream dependencies. It is the AI counterpart to a software bill of materials and the prerequisite for governing or attesting to AI.
How do I build an AI-BOM?
Enumerate the models in use, catalog the agents and the tools they can call, map non-human identities to their owners, trace the supply-chain dependencies, and then reconcile the inventory against what is actually acting so shadow AI surfaces.
How does an AI-BOM relate to shadow AI?
The AI-BOM is the list of AI you sanctioned; shadow AI is what is acting but not on the list. Comparing observed non-human actors against the BOM is exactly how ungoverned agents and MCP servers surface.
Why do I need an AI-BOM for compliance?
Because attesting to an AI system, answering a vendor security review, or mapping controls all require knowing the system’s parts. The AI-BOM is the inventory those activities depend on.
Keep reading
Guides
Calculators
One finance briefing a week
New guides, checklists and benchmarks, by email. Unsubscribe from any issue.
Inventory your AI, then govern it
Build the living AI-BOM governance depends on. Talk to us: we will map it to your stack.