Privacy Policy

Overview & Privacy Governance Roles

Pallasite AI provides autonomous refund governance, fraud detection, and policy abuse prevention software to e-commerce merchants ("Merchants"). This Privacy Policy details how data is ingested, processed, protected, and purged when Merchants install Pallasite AI on their Shopify stores or interact with our API suite.

Information We Collect & Process

We adhere strictly to the principle of data minimization, collecting only data required to perform fraud and policy verification:

• Merchant Store Profile: Shopify store domain, business email, store currency, timezone, and user authentication records.
• Refund & Return Metadata: Shopify order ID, line item SKUs, purchase timestamps, return reason strings, package weight manifests, and carrier tracking identifiers stored for dispute defense and policy auditing.
• Customer PII Handling & Ephemeral Model Ingestion: Customer claim text and dispute notes are evaluated in volatile RAM for stateless model reasoning with automated client-side PII scrubbing. Customer names and emails are never used to train AI models.
• Encrypted BYOK Credentials: Merchant-provided API keys for supported providers (Google Gemini, OpenAI, OpenRouter, and Anthropic Claude), encrypted at rest using AES-256-GCM with per-tenant encryption keys.
• Forensic Audit Records & Data Retention: Generated risk scores, confidence metrics, heuristic rule matches, and cryptographic SHA-256 Merkle fingerprints are persisted in encrypted database storage during your active merchant tenure to provide auditable proof for payment disputes. Historical audit records and buyer identifiers are scrubbed upon receiving Shopify GDPR/CCPA erasure requests (within 30 days) or purged within 48 hours following app uninstallation.

Shopify Mandatory Privacy Webhooks Compliance

In strict compliance with Shopify App Store Partner specifications, Pallasite AI natively processes mandatory GDPR/CCPA webhooks:

Zero Model-Training Guarantee

Pallasite AI utilizes frontier Large Language Models (Google Gemini 3.8 Flash, OpenAI GPT-6 Astra [Daybreak/Enterprise API preview; falling back to Gemini 4 Argon when unavailable], DeepSeek V4 Pro, GLM-5.3 Fast/Flash & Anthropic Claude Opus 5.5 / Claude Sonnet 5.5). Under our Bring-Your-Own-Key (BYOK) architecture, inference requests are sent directly to provider APIs according to standard commercial API terms prohibiting model training on API payloads:

Sub-Processors & Cloud Infrastructure

We utilize trusted cloud hosting and model provider infrastructure. Under our Bring Your Own Key (BYOK) architecture, AI inference calls are transmitted to the respective providers under the merchant's direct provider accounts and commercial API privacy settings:

• Render Services Inc.: Application container orchestration, edge compute, TLS termination, and API routing infrastructure.
• Neon Inc.: Encrypted serverless PostgreSQL database infrastructure with automated backups and AES-256 storage encryption.
• Google LLC (Google Cloud / Google AI Studio): Primary triage model inference (Gemini 3.8 Flash / Gemini 4 Argon) executed via merchant-configured BYOK API keys under Google AI API commercial terms.
• Anthropic PBC: Consensus reasoning model inference (Claude Opus 5.5 / Claude Sonnet 5.5) executed via merchant BYOK credentials under commercial API privacy terms.
• OpenAI LLC: Flagship fraud audit reasoning (GPT-6 Astra [Daybreak/Enterprise API preview] / GPT-5.6 Terra) executed via merchant BYOK credentials under OpenAI API data privacy terms.
• OpenRouter Inc.: Routing infrastructure for open-weights and specialized consensus models, routing inference directly to upstream model providers including DeepSeek (DeepSeek V4.1 Flash) and Zhipu AI (GLM-5.3 Flash) as named upstream recipients under merchant BYOK credentials.

Security Controls & Data Safeguards

• Encryption at Rest: All stored database records and credentials are protected by AES-256-GCM encryption.
• Encryption in Transit: 100% of incoming webhooks, dashboard sessions, and API traffic require TLS 1.3 encryption.
• Role-Based Access Control (RBAC): Strict tenant isolation enforced at the database query layer ensures zero cross-merchant data leakage.
• Automated Token Scrubber: Sensitive customer PII (credit card fragments, SSNs, personal phone numbers) is automatically masked prior to model inference.