SecureAI Security Architecture

SecureAI's security architecture is designed to protect automotive aftermarket data at every layer — from the browser to the database to external AI model providers. This article explains the technical security controls, encryption standards, storage architecture, and data flow protections that secure your organization's information.

Architecture Overview

SecureAI runs on Google Cloud Platform (GCP) Cloud Run and follows a layered security model:

┌─────────────────────────────────────────────────┐
│  Users (Browser / API Clients)                  │
│  ── TLS 1.2+ ──────────────────────────────     │
├─────────────────────────────────────────────────┤
│  Cloud Run Frontend (HTTPS-only)                │
│  ── mTLS ───────────────────────────────────    │
├─────────────────────────────────────────────────┤
│  API Server (Authentication + Authorization)    │
│  ── mTLS ───────────────────────────────────    │
├─────────────────────────────────────────────────┤
│  Data Layer            │  Model Providers       │
│  (Encrypted at rest)   │  (TLS, no retention)   │
└─────────────────────────────────────────────────┘

Each layer enforces its own security boundary. A compromise at one layer does not automatically grant access to another.

Encryption in Transit

All data moving between components is encrypted:

TLS Configuration

• Cipher suites: Only strong cipher suites are permitted. Weak ciphers (RC4, DES, 3DES) and protocols (SSLv3, TLS 1.0, TLS 1.1) are disabled.
• Certificate management: TLS certificates are managed automatically through GCP's certificate infrastructure. Certificates are rotated before expiration without manual intervention.
• HSTS: HTTP Strict Transport Security headers are set with a minimum max-age of one year, preventing protocol downgrade attacks.

Encryption at Rest

All stored data is encrypted using AES-256:

Customer-Managed Encryption Keys (CMEK)

Organizations with stricter key management requirements can use CMEK through GCP's Key Management Service (KMS):

• How CMEK works: Your organization controls the encryption keys used to protect your data. GCP KMS stores and manages the keys, but your organization controls their lifecycle — including creation, rotation, and revocation.
• Key rotation: Automatic key rotation is configured by default (every 90 days). Your organization can adjust the rotation period or trigger manual rotation through GCP KMS.
• Key revocation: Revoking your CMEK renders all data encrypted with that key permanently inaccessible. Use this as an emergency measure only.
• Availability: CMEK is available for enterprise service agreements. Contact your account representative during onboarding to configure CMEK.

Storage Architecture

Data Storage Locations

SecureAI stores data in three primary locations:

1. Cloud SQL (PostgreSQL)

• Stores conversation history, user accounts, system configuration, and audit logs.
• Runs in a high-availability configuration with automatic failover.
• Point-in-time recovery enabled, with backups retained according to your service agreement.
• Database access restricted to the API server — no direct external connections are permitted.

2. Cloud Storage

• Stores uploaded documents (PDFs, images, spreadsheets) and processed document embeddings.
• Objects are organized by organization ID, enforcing tenant isolation at the storage level.
• Lifecycle policies automatically delete objects according to your data retention configuration.
• Versioning is disabled by default to prevent unintended data retention. Deleted files are permanently removed.

3. In-Memory Processing

• AI model requests and responses are processed in memory and are not persisted to disk.
• Temporary data (file parsing buffers, embedding generation intermediates) exists only for the duration of the request.
• Cloud Run instances are ephemeral — when an instance scales down, all in-memory data is destroyed.

Data Residency

Data storage is region-specific:

All stored data — conversations, documents, user accounts, audit logs, and backups — resides in the selected region. Changing regions after deployment requires a planned migration coordinated with your account representative.

Model Provider Data Handling

When SecureAI sends a prompt to an AI model provider, specific protections apply:

What Is Sent to Model Providers

• The user's prompt (question or instruction).
• Relevant conversation context (prior messages in the current conversation, subject to the model's context window).
• Retrieved document content when knowledge base or RAG features are used.
• System instructions configured by your organization's administrator.

What Is NOT Sent to Model Providers

• User account credentials or authentication tokens.
• Data from other users' conversations or other organizations.
• Audit log content.
• Administrative configuration details.

Provider Data Protection Guarantees

Supported Model Providers

SecureAI supports multiple model providers. Each provider's data handling is governed by SecureAI's data processing agreements:

• OpenAI: API requests processed under OpenAI's enterprise API terms (zero data retention).
• Anthropic: API requests processed under Anthropic's commercial API terms (zero data retention).
• Azure OpenAI: Requests processed within your organization's Azure tenant when configured (data stays in your Azure environment).
• Local models (Ollama, vLLM): When configured, prompts are processed on infrastructure you control. No data leaves your environment.

Your administrator selects which providers are available. See Adding Custom Model Providers for configuration details.

Network Security

Perimeter Controls

• Cloud Run ingress: Configured to accept traffic only from authorized sources. Internal services are not exposed to the public internet.
• IP allowlisting: Organizations can restrict access to specific IP ranges. See Setting Up IP Allowlisting for Enterprise Access.
• DDoS protection: GCP Cloud Armor provides automatic DDoS mitigation for all incoming traffic.
• WAF rules: Web Application Firewall rules block common attack patterns (SQL injection, XSS, path traversal) at the edge.

Internal Network

• All internal service communication uses mTLS.
• Database access is restricted to the API server via private networking (VPC). No public IP is assigned to the database.
• Secrets (API keys, database credentials, encryption keys) are stored in GCP Secret Manager, not in application code or environment variables.

Authentication and Authorization

Authentication

SecureAI supports multiple authentication methods:

Authorization

• Role-based access control (RBAC): Users are assigned roles (User, Admin) that determine access levels.
• Tenant isolation: All authorization checks include the organization ID. A user in one organization cannot access another organization's resources regardless of role.
• API scope enforcement: API tokens are limited to the permissions granted at creation. Tokens cannot exceed the creating user's permissions.

Vulnerability Management

• Dependency scanning: Automated scanning of application dependencies for known vulnerabilities (CVEs). Critical vulnerabilities are patched within 48 hours of disclosure.
• Container image scanning: Cloud Run container images are scanned before deployment. Images with critical or high-severity vulnerabilities are blocked from deployment.
• Penetration testing: Independent security assessments are conducted regularly. Summary reports are available under NDA — contact your account representative.
• Incident response: Defined incident response procedures with documented escalation paths. Security incidents are communicated to affected organizations according to contractual SLAs and regulatory requirements.

Security Monitoring

• Audit logging: All administrative and security-relevant actions are logged with timestamps, user identity, source IP, and action details.
• Anomaly detection: Automated alerting on suspicious patterns — repeated authentication failures, unusual data access volumes, configuration changes outside business hours.
• Log retention: Security logs are retained independently of conversation data, typically for a longer period as defined in your service agreement.
• SIEM integration: Audit logs can be exported for integration with your organization's security information and event management (SIEM) tools.

Frequently Asked Questions

Is my data encrypted?

Yes. All data is encrypted both in transit (TLS 1.2+) and at rest (AES-256). Organizations requiring additional key control can use Customer-Managed Encryption Keys (CMEK).

Can AI model providers see my data?

Model providers process your prompts to generate responses, but they cannot retain, store, or use your data for training. This is enforced by SecureAI's data processing agreements with each provider.

Where is my data physically stored?

By default, in GCP's us-central1 region (United States). EU data residency (europe-west1, Belgium) is available on request. Additional regions are available for enterprise agreements.

Can I use my own encryption keys?

Yes. CMEK support is available for enterprise service agreements. Your organization controls key lifecycle through GCP's Key Management Service.

How does SecureAI prevent cross-tenant data access?

Tenant isolation is enforced at multiple layers — application-level authorization checks, database-level row isolation, and storage-level object organization by organization ID. All queries include organization scope, and there is no mechanism for cross-tenant access.

Does SecureAI store data on local devices?

No. All data is stored server-side in GCP. No conversation data, documents, or credentials are persisted on end-user devices.

Related Articles

• How SecureAI Handles Your Data
• Compliance Certifications — SOC 2, GDPR, HIPAA
• Setting Up IP Allowlisting for Enterprise Access
• Configuring Data Retention Policies
• Adding Custom Model Providers
• How to Configure SAML SSO

Questions

For security architecture questions, to request penetration test reports, or to discuss CMEK configuration, contact your account representative.