Interactive web tools for AI research oversight and IRB review, built with SvelteKit 5.

This project provides web-based tools to help IRBs, researchers, and developers:

• Identify and assess AI-specific risks in human subjects research
• Navigate the 3-Stage IRB Review Framework
• Generate protocol requirements and mitigation strategies
• Trace connections between risks, mitigations, controls, and regulations

All risk content is derived from the SME-reviewed AIHSR Risk Reference Tool v1.5 by Tamiko Eto, augmented with the MIT AI Risk Repository mitigation database.

https://malathon.github.io/ai-oversight-tools/

The application uses Graphology as the single source of truth for all entity relationships. The directed graph contains:

1. Explicit Links - From traceability data (trigger, mitigation, control, regulation relationships)
2. Question Dependencies - showIf conditions as dependency edges
3. Control-Subcategory - contains edges linking controls to their mitigation strategies

All derived data (link counts, connections, dependencies) are computed as cached graph queries using Svelte 5's $derived.by() reactive system.

┌────────────────────────────────────────────────────────────────────┐
│                      traceGraph (Graphology)                       │
│  ┌─────────┐    trigger    ┌──────┐   mitigation   ┌────────────┐ │
│  │Question │──────────────▶│ Risk │──────────────▶│ Subcategory │ │
│  └─────────┘               └──────┘               └────────────┘  │
│       │                        │                       │          │
│       │ dependency             │ regulation            │ contains │
│       ▼                        ▼                       ▼          │
│  ┌─────────┐              ┌──────────┐            ┌─────────┐     │
│  │Question │              │Regulation│            │ Control │     │
│  └─────────┘              └──────────┘            └─────────┘     │
└────────────────────────────────────────────────────────────────────┘

The system collects stage-specific guidance during graph traversal:

• Risks: stageGuidance - risk context per stage
• Subcategories: stageGuidance + stageAppropriateness - strategy guidance + importance level
• Controls: implementationNotes - implementation guidance per stage

This enables LLM synthesis by collecting all relevant guidance along traversal paths from any starting node.

ai-oversight-tools/
├── webapp/                         # SvelteKit 5 application
│   ├── src/
│   │   ├── routes/
│   │   │   ├── admin/             # Traceability graph editor
│   │   │   ├── innovator/         # Innovator self-assessment
│   │   │   ├── reviewer/          # Reviewer checklist
│   │   │   └── risk-matrix/       # Risk assessment matrix
│   │   └── lib/
│   │       └── admin.ts           # Shared filters and utilities
│   └── static/data/               # JSON data files
│       ├── assessment-questions.json
│       ├── risk-domains.json
│       ├── risk-subdomains.json
│       ├── mitigation-strategies.json
│       ├── technical-controls.json  # 600+ controls from MIT AI Risk Repository
│       ├── traceability.json        # Entity linkages
│       └── unified-schema.json
│
├── packages/rules-data/            # Legacy data (deprecated)
│
└── docs/
    ├── reference/aihsr/            # AIHSR Risk Reference Tool source data
    └── planning/                   # Implementation plans

• Framework: SvelteKit 5 with Svelte 5 runes ($state, $derived, $effect)
• Graph Library: Graphology for directed graph operations
• Build: Vite 7
• Deployment: GitHub Pages via GitHub Actions

cd webapp
npm install
npm run dev

Build and check:

npm run build
npm run check

AIHSR Risk Reference Tool © 2025 by Tamiko Eto Licensed under CC BY-NC-SA 4.0

MIT AI Risk Repository data used under MIT license.

• AIHSR Risk Reference Tool: Tamiko Eto, MA CIP (TechInHSR.com)
• 3-Stage IRB Review Framework: Eto, Lifson, Vidal (Frontiers in Systems Biology, 2026)
• MIT AI Risk Repository: Primary source for risk taxonomy and technical controls

• Email: [email protected]
• Scheduling: https://calendly.com/etohtamiko
• Feedback: https://forms.gle/2ij9ic3N8ePN1i799