Multi-signal discovery
Network egress to model providers, SaaS administration data, expense records, code and configuration, and structured interviews.
Warden · AI Governance
Every AI governance program begins with a list nobody has
A discovery engagement that finds the models, agents, copilots and embedded AI features actually in use across your organization, and turns them into a maintained inventory rather than a one-off spreadsheet.
AI arrives through the side door
Most AI in an enterprise was not procured as AI. It arrived as a feature in a SaaS product already under contract, a developer tool paid for on a card, or a copilot enabled by default in a suite you already own.
- Embedded features are invisible to procurementThe vendor added AI to a product you bought three years ago and told you in a release note.
- Self-declaration undercountsAsking teams what AI they use reliably returns a fraction of what network and expense data show.
- A static inventory decays immediatelyWithout a maintenance mechanism, the list is wrong within a quarter and worse than useless at audit.
How the engagement runs
Classification
Purpose, data touched, autonomy level, role held and business criticality per system.
Ownership assignment
A named accountable owner per entry, because an inventory without owners cannot be maintained.
Maintenance mechanism
Connectors and an intake process so new AI use registers itself rather than being rediscovered next year.
What you hold at the end
- A classified inventory of AI systems, agents and embedded AI features in use
- Data-touched and autonomy classification per entry, ready for regulatory classification
- A named owner for every entry
- A running discovery and intake mechanism so the inventory stays current
- A shadow-AI findings summary for the risk committee
What keeps it current
You cannot govern what you have not found
Discovery is the cheapest engagement here and the one that changes the plan most often.