AI-Control Architecture¶

Library AreaTemplate Library

Current PageAI-Control Architecture

FocusAI boundaries and control

UseDefine AI use, review, prohibitions, and evidence

Repository Areadocs/template-library/architecture/

← Previous Quality Attribute Strategy

Current AI-Control Architecture

Next → Architecture Decision Records

Template purpose¶

Use this template to document AI use and AI control.

If AI is not used, say so clearly. If AI is used, describe its purpose, inputs, outputs, boundaries, review points, prohibited behaviors, failure handling, evidence, evaluation, monitoring, and human accountability.

Project¶

<Project name>

Document control¶

Field	Value
Artifact owner	`<owner>`
Primary reviewers	`<reviewers>`
Status	`<draft / in review / accepted / revised>`
Last updated	`<YYYY-MM-DD>`
Related Engineering Stage	`ES-104 — Architecture and Technical Approach`
Project workspace target	`docs/project-workspace/architecture/ai_control_architecture.md`

AI use status¶

Select one:

AI is not used in this architecture.

or

AI is used in limited, controlled ways described below.

AI capability register¶

AI Capability	Purpose	Inputs	Outputs	Human Review	Evidence Produced	Status
`<capability>`	`<purpose>`	`<inputs>`	`<outputs>`	`<review>`	`<evidence>`	`<planned / allowed / limited / deferred / prohibited>`

AI role classification¶

Capability	AI Role	Decision Authority	Human Accountability
`<capability>`	`<assist / summarize / classify / recommend / generate / route / monitor / none>`	`<AI proposes / human decides / automated within boundary / prohibited>`	`<owner>`

AI boundaries¶

Boundary	Description	Control
`<boundary>`	`<description>`	`<control>`

Human review points¶

Review Point	Reviewer	Review Trigger	Evidence Produced
`<review point>`	`<reviewer>`	`<trigger>`	`<evidence>`

AI prohibitions¶

Prohibition	Rationale	Enforcement / Verification
`<prohibition>`	`<rationale>`	`<control, test, review, monitoring>`

Input controls¶

Input Type	Control	Risk Addressed
`<input>`	`<validation, filtering, access control, minimization, approval>`	`<risk>`

Output controls¶

Output Type	Control	Risk Addressed
`<output>`	`<review, labeling, confidence handling, explanation, blocking, escalation>`	`<risk>`

Failure handling¶

<What happens when AI output is missing, wrong, unsafe, biased, inconsistent, unavailable, or out of policy?>

Escalation and override¶

Scenario	Escalation Path	Override Authority
`<scenario>`	`<path>`	`<role>`

AI evidence¶

<How AI use is recorded and reviewed.>

Evidence	Created By	Stored In	Retention / Review
`<prompt/output log, AI use log, review record, evaluation record, escalation record>`	`<component / human / process>`	`<location>`	`<rule>`

Evaluation and monitoring¶

Capability	Evaluation Method	Monitoring Signal	Failure Threshold
`<capability>`	`<method>`	`<signal>`	`<threshold>`

Model, vendor, or tool dependency¶

Dependency	Risk	Mitigation
`<model, tool, service, vendor, library>`	`<risk>`	`<mitigation>`

Review checklist¶

[ ] AI use status is explicit.
[ ] AI capabilities are listed or marked not applicable.
[ ] Human accountability is defined.
[ ] AI boundaries and prohibitions are explicit.
[ ] Input and output controls are defined.
[ ] Failure handling is defined.
[ ] AI evidence is defined.
[ ] Evaluation and monitoring are addressed.
[ ] Escalation and override paths are defined.

Continue to Architecture Decision Records

Record the major decisions, alternatives, rationale, consequences, and tradeoffs.

Open ADRs →