Orcana.AI Data Privacy & Security

Last Updated: September 30, 2025
Effective Date: August 14, 2024

We design every system, model, and workflow to ensure that no raw customer data ever leaves your environment unprotected. This document outlines how Orcana’s architecture, encryption standards, and privacy controls uphold the highest standards of enterprise data protection.

Guiding Principle: Zero Raw-Data Exposure

Orcana ensures that sensitive customer data is never directly processed, stored, or viewed by Large Language Models (LLMs) or any third-party service. Our product architecture separates data ingestion, processing, and interpretation layers to maintain total data isolation and integrity.

• No external API exposure — all model interactions happen through Orcana-managed APIs.
• LLMs never see raw data — all customer identifiers are encrypted or replaced with secure placeholders.
• No training on client data — neither Orcana nor Azure OpenAI uses your data for model fine-tuning.
• Traceable and explainable outputs — every analytic result is verifiable and auditable through Orcanalytics™.

System Architecture Overview [See Appendix for diagram]

Orcana’s architecture follows a multi-layered secure analytics pipeline designed for accuracy, privacy, and compliance.

Data Flow & Analysis

Data Ingestion — Data is pulled securely from client CRMs or data warehouses (e.g., Snowflake) through API integrations.

Data Staging — Incoming data is cleaned, normalized, and stored temporarily in a staging environment for pre-processing.

Analytics Execution — Orcana’s proprietary AI engine, Orcanalytics™, performs deterministic analysis on staged data — no data leaves the environment unencrypted.

Insight Generation — The natural language layer converts analytic output into readable insights while preserving full data anonymization.

LLM Integration & Secure AI Usage

3.1 Controlled Model Interaction

Orcana leverages Azure OpenAI for natural language generation but operates through independent, Orcana-owned API gateways.

Each request to the LLM:

• Passes through Orcana’s encryption and validation layer.
• Uses encrypted entity placeholders (e.g., Product_ID_101 instead of “Drug A”).
• Returns structured text responses that are decrypted and validated internally.

3.2 Secure Azure Deployment

• All LLM calls are routed through Azure OpenAI deployments bound to Orcana’s tenant and region.
• Data isolation: Azure guarantees that prompts and completions are never stored or used for model retraining.
• Dedicated access control: Only Orcana systems can communicate with these deployments.

Deterministic Analytics with Orcanalytics™

Unlike generic AI wrappers, Orcana never uses LLMs to perform analytics. Our proprietary engine, Orcanalytics™, executes all data retrieval, analysis, and visualization logic deterministically — ensuring accuracy, repeatability, and compliance.

Key safeguards:

• No code generation on the fly — all analytics logic is written and maintained in-house.
• Industry- and domain-specific analytics modules ensure contextual accuracy.
• Ground-truth validation modules compare every LLM summary to the underlying analytic result to prevent hallucinations.
• All analytics pipelines are fully auditable and version-controlled.

Data Management and Processing

5.1 Data Minimization

Orcana integrates with client data warehouses using secure API connections and extracts only the fields necessary for analysis. This ensures:

• No unnecessary or redundant data transfer.
• Adherence to data minimization principles under GDPR and similar frameworks.

5.2 PII Redaction

Before any processing or model interaction:

• All Personally Identifiable Information (PII) is redacted using algorithmic redaction techniques.
• Optional “mapping IDs” allow clients to reconcile analytics with internal datasets while keeping Orcana’s systems anonymized.

Aggregation & Summarization

For analytics requiring trend or performance metrics, Orcana generates aggregated summaries instead of storing identifiable records.

This ensures that no individual or entity can be traced from stored data.

Data Storage and Encryption

Hosting:
Orcana is fully deployed on Microsoft Azure, utilizing its enterprise-grade data storage infrastructure (Azure Blob, File, and SQL as applicable).

Encryption:

• AES-256 encryption at rest
• TLS 1.2+ encryption in transit

Access Control:

• Role-based access control (RBAC) and least-privilege policies.
• All service access requires secure authentication and audit logging.

Data Refresh:

• Stored data is refreshed on a defined cadence to maintain accuracy and compliance.
• No residual raw data persists beyond operational necessity.

Security Layer & Workflow Validation

Five-Step Secure Processing Workflow

Named Entity Encryption (Python): Sensitive identifiers (products, accounts, HCPs, reps) are encrypted.

Context Ingestion: Encrypted data and business rules are bundled securely.

LLM Response Generation: The LLM sees only anonymized placeholders.

Query → Result Validation: The generated response is validated against the analytical truth.

Decryption & Delivery: The encrypted entities are mapped back to real terms, and insights are presented securely.

Each output passes through relevance and consistency validation layers before being shared, guaranteeing accuracy and data safety.

Security Principles

Orcana’s systems are built around the CIA Triad — Confidentiality, Integrity, and Availability.

Confidentiality: No external access to customer data; encrypted throughout lifecycle.
Integrity: All workflows and outputs are logged, validated, and tamper-proof.
Availability: Highly available Azure infrastructure with redundancy and disaster recovery.

Additional safeguards include:

• Continuous vulnerability monitoring.
• SOC 2–aligned operational controls.
• Annual penetration testing by independent assessors.
• Role-based permissions enforced at every layer.

Customer Trust & Compliance

Orcana’s privacy and data management practices align with major global frameworks, including:

• GDPR (EU)
• CCPA (California)
• HIPAA readiness (for de-identified health data)
• SOC 2 Type II control alignment

All client data handling practices are transparent and can be reviewed under mutual NDA for compliance assurance.

Why Orcana Thrives on Security

• No raw data ever touches LLMs.
• All interactions occur within Orcana’s secure Azure environment.
• Outputs are deterministic, verifiable, and explainable.
• Every insight is traceable back to its analytic source.
• Customers maintain full ownership and visibility of their data at all times.

Contact Information

Orcana-AI, Inc. 1111B S Governors Ave STE 21776 Dover, DE, 19904 US 📧 info@orcana.ai

Appendix

System architecture diagram