AI Governance & Transparency

This document describes how Inori uses artificial intelligence. It is provided for transparency and compliance with applicable AI governance regulations, including the Colorado AI Act (SB 24-205).

Status: DRAFT — Pending legal review before publication.

---

1. System Purpose

Inori uses AI to extract structured data from insurance Certificates of Insurance (COIs) and compare the extracted data against user-configured compliance requirements. The system assists compliance review — it does not make final compliance determinations.

2. AI Technology

• Model: Anthropic Claude (Haiku for standard extraction, Sonnet for complex documents requiring escalation)
• Method: Computer vision + structured extraction from PDF documents
• Scope: ACORD 25 certificate forms (89 fields extracted across 10 sections)
• Processing: Ephemeral — documents are sent to the AI provider for extraction only. Documents are not used for model training.

3. Data Inputs

The AI system processes:

• Certificate of Insurance PDF documents uploaded by users or vendors
• No personal health information (PHI), financial account numbers, or social security numbers are processed
• Extracted structured data (coverage limits, dates, names, endorsements) is stored in the platform database

4. Testing & Accuracy

• Calibration suite: 273 test scenarios across 19 categories, verified at 100% accuracy
• Accuracy claim: Greater than 95% on standard ACORD 25 certificate forms
• Known limitations:
  • Low-quality scans, handwritten, or faded documents may produce reduced accuracy
  • Non-standard certificate formats (carrier-specific forms) may have lower extraction rates
  • Multi-page continuation certificates may miss fields on subsequent pages
  • Endorsement verification is text-based (form code detection), not endorsement document validation

5. Human Oversight

• All AI compliance results are presented as recommendations, not final determinations
• Users can manually review and override any AI-extracted field via the review system
• Users can approve certificates "as-is" or edit individual fields with audit trail
• The consent gate (shown on first use) discloses the AI nature of the analysis
• Every result displays: "AI-analyzed · Review recommended for binding decisions"

6. Consumer Notice & Rights

Notice

• A consent dialog is displayed before any AI analysis results are shown
• All compliance results include an inline AI disclosure marker
• Email notifications include: "Certificate analysis powered by AI. Not insurance advice."

Appeal & Correction

• Vendors who disagree with compliance results can upload a corrected certificate at any time via the vendor portal
• Certificate holders can request manual review by their property manager
• The manual review system allows field-by-field correction with audit trail

Opt-Out

• Users can choose not to use AI-powered analysis and perform manual compliance review
• The platform does not make automated decisions without human review capability

7. Record Retention

• AI extraction results are stored for the lifetime of the user's account
• Deleted data is soft-deleted and permanently purged 90 days after account closure
• Audit trails of AI analysis, manual reviews, and compliance decisions are retained for compliance purposes

8. Risk Classification

Under the Colorado AI Act (SB 24-205), Inori's AI system may be classified as a "high-risk" system if compliance determinations result in consequential decisions affecting vendor access to projects. Inori mitigates this through:

• Human-in-the-loop design (all results require user action)
• Appeal mechanisms (vendor re-upload, manual review)
• Consumer notice (consent gate, inline markers)
• Documentation of testing methodology and accuracy

9. Contact

For questions about Inori's AI governance: ask@askinori.com