Reference architecture
The reference implementation: AI, data and security as one system
By Richard White ·
Free, no form.
Abstract
The implementation companion to The Four-Part System. Production AI fails at the seams between AI, data and security programs. This twenty-two-page paper shows one way to build and operate the three as a single system, naming real products as one example combination: the layers, the data platform, business meaning and knowledge, interfaces, and AI and agents.
Four sections cover security: labels and tags that follow data from classification to every interface, agents treated like employees with a registered identity, a role, a manager and a defined exit, an observability control plane that brings every agent action to the security operations center, and one owner for each control. Orchestration and the operating model hold the pieces together.
Two illustrative use cases are traced end to end, step by step: anti-money-laundering alert investigation at a US bank and claims handling at a US insurer. It closes with an implementation path and a readiness checklist. Products are named as examples of each pattern, not endorsements, and no vendor reviewed or paid for the paper.
Decision it supports
Which architecture dependencies and failure paths must support the first read-only release.
What you will produce
A dependency register tied to a workflow, tested scope boundary and release evidence.
See it worked through in a guide
Claims package 1.1.0 · 26 October 2026. Synthetic worked example: team A, group disability, region 2. Read-only preparation of the assigned case; examiners retain every case decision. Hold candidate 1.0.0 after an unrelated employee’s claim is returned. In the synthetic decision, accept 1.1.0 for a monitored team A rollout only after prior-claims history is disabled and the remaining scope passes a fresh evaluation. All figures, evidence and sign-offs in that guide are invented; the paper provides the underlying method.
In the paper
Three passages, as published
Classification is not authorization
A DSPM tool informs the controls that govern each surface. It does not replace them, and it should not become a second, competing control.
Agent identity and supervision
Use the employee lifecycle as a starting pattern. Give each agent a registered identity, an approved purpose, defined permissions, monitoring and a retirement process.
One owner per control
No component should silently recreate another component’s decision.
Next step
Discuss this paper against your own program.
A free sixty-minute working session with Richard on your top gaps. You leave with a one-page summary within three business days, and three next questions.