Skip to content

AI-Control Architecture

LMU/COICP Example

AI-Control Architecture Example

Define how LMU/COICP keeps AI Incident Summary deferred/disabled for RC-001 while preserving future AI controls for scenarios like Damen Hall water leak summaries: draft-only output, human review, labeling, evidence, and governance.

ES-104 AI Control Disabled for RC-001
Example AreaLMU/COICP
Current PageAI-Control Architecture
FocusLMU AI boundaries
UseControl deferred AI assistance
Repository Areadocs/examples/lmu-coicp/architecture/
← Previous Quality Attribute Strategy
Current AI-Control Architecture
Next → Architecture Decision Records

Example purpose

This artifact defines AI-control architecture for LMU/COICP.

AI is not needed to prove the first baseline. The architecture intentionally keeps AI Incident Summary disabled so core incident workflow, evidence history, access control, and emergency-boundary behavior can be verified first.

Project

LMU Campus Operations and Incident Coordination Platform

Document control

Field Value
Artifact owner AI reviewer
Primary reviewers Architecture Review Board chair, COICP Product Engineer, IT security reviewer, Compliance reviewer
Status Accepted with AI Incident Summary deferred/disabled
Last updated 2026-07-06
Related Engineering Stage ES-104 — Architecture
Project workspace target docs/project-workspace/architecture/ai_control_architecture.md

Architecture context

Field Value
Source planning package ES-103 — Planning
Architecture baseline produced LMU-COICP-ARCH-001
Design baseline expected next LMU-COICP-DES-001
Implementation baseline expected later LMU-COICP-IMPL-001
Integrated baseline expected later LMU-COICP-INT-001
Release candidate expected later LMU-COICP-RC-001
Downstream release posture Internal engineering review only
Planned downstream review environment LMU-COICP Internal Engineering Review Environment
Planned downstream dataset Spring Semester Synthetic Incident Dataset
Planned downstream synthetic records 421
Planned downstream reviewer accounts 24
Planned downstream review window March 18–22, 2026
AI Incident Summary Deferred / disabled
Operational pilot Not approved

Current AI status

AI Incident Summary is disabled for LMU-COICP-ARCH-001 and must remain disabled through LMU-COICP-RC-001 unless a future readiness cycle explicitly approves it.

Current scenario behavior

Scenario AI Behavior
COICP-SYN-001 no AI summary available
COICP-SYN-118 no AI draft of Damen Hall handoff summary
COICP-SYN-204 no AI access to residence scenario
COICP-SYN-219 no AI closure summary
COICP-SYN-322 no AI emergency guidance
COICP-SYN-407 no AI Wi-Fi disruption summary

Future AI extension point

If AI is enabled later for a scenario such as COICP-SYN-118, future flow would be:

Authorized reviewer
  → Feature check confirms future AI approved
  → Selected evidence timeline sent to AI Assistance Adapter
  → AI returns draft only
  → Human reviewer edits / accepts / rejects
  → Human Review Service records decision
  → Evidence History Service records accepted human action

AI components

Component Current RC-001 Status Future Responsibility
AI Assistance Adapter disabled generate draft summary from approved context
Human Review Service disabled require accept/edit/reject before official use
Evidence History Service active future AI request/review evidence if enabled
Configuration / Policy Layer active keep AI feature disabled
Access Control Service active deny AI summary requests

AI prohibitions

  • no AI summary generation in RC-001;
  • no AI emergency advice for COICP-SYN-322;
  • no AI official closure for COICP-SYN-219;
  • no AI owner assignment for COICP-SYN-118;
  • no AI access to COICP-SYN-204 unless future scope and role allow it;
  • no hidden AI-generated content;
  • no AI-generated official evidence without human acceptance;
  • no AI enablement without design, implementation, integration, testing, release, operations, deployment, monitoring, stewardship, and governance evidence.

AI evidence if enabled later

Future AI use would require:

  • request event;
  • requesting user;
  • scenario / incident reference;
  • selected context reference;
  • draft marker;
  • generated timestamp;
  • human acceptance / edit / rejection action;
  • accepted text if any;
  • retention disposition;
  • AI-use monitoring signal.

Design handoff

Design must define:

  • AI-disabled UI behavior;
  • AI feature check interface;
  • future AI draft data structure;
  • future HumanReviewAction record;
  • denial behavior for all AI summary requests in RC-001;
  • prohibited-data rules for any future AI context.

Continue to Architecture Decision Records

Record scenario-driven architecture decisions that later Design implements.

Open Architecture Decision Records →