Security Architecture
Kanawai AI is built from the ground up with enterprise-grade security. This document details our authentication controls, data isolation, encryption practices, AI data privacy guarantees, compliance certifications, and the security inherited from Google Cloud Platform.
Authentication & Access Control
Kanawai AI enforces rigorous authentication controls to ensure that only authorized users access the platform. Every account is protected by multiple layers of identity verification.
Password & MFA Requirements
- Minimum 12 alphanumeric characters required for all user passwords, enforcing a strong password baseline that exceeds NIST SP 800-63B recommendations.
- Multi-Factor Authentication (MFA) is mandatory for all user accounts. MFA cannot be disabled at the user level — it is enforced organization-wide as a platform requirement.
- Passkey support enables passwordless authentication using FIDO2/WebAuthn standards, allowing biometric or hardware security key authentication for enhanced phishing resistance.
Enterprise Identity Federation
- OAuth 2.0 integration enables Single Sign-On (SSO) with enterprise identity providers including Google Workspace, Microsoft Entra ID, and Okta.
- SAML 2.0 federation for organizations requiring standards-based SSO with their existing identity infrastructure, including support for just-in-time (JIT) user provisioning.
- Session management with configurable idle timeouts, automatic session expiration, and concurrent session limits per user. All session tokens are cryptographically signed and bound to device context.
Data Isolation by Geographic Region
Kanawai AI provides strict tenant isolation with regional data residency controls deployed on Google Cloud Platform. Customer data never leaves the geographic boundary selected during tenant provisioning.
United States
us-central1, us-east1, us-west1
Europe
europe-west1, europe-west3, europe-north1
Asia-Pacific
asia-southeast1, asia-northeast1
Tenant Isolation Architecture
- Dedicated project-level isolation — each customer tenant operates within a segregated GCP project with independent IAM policies, networking controls, and encryption keys.
- VPC Service Controls create security perimeters around GCP resources, preventing data exfiltration and ensuring that API calls are restricted to authorized network boundaries.
- Network segmentation with private IP addressing, Cloud NAT for controlled egress, and firewall rules that enforce least-privilege network access between services.
- Data residency guarantees enforced through GCP Organization Policies that restrict resource creation to the customer's selected geographic region, preventing accidental or unauthorized data migration.
Customer Managed Encryption Keys (CMEK)
Kanawai AI supports Customer Managed Encryption Keys (CMEK) via Google Cloud Key Management Service (KMS), providing customers with full cryptographic control over their data at rest.
How CMEK Works
- Customer-owned encryption keys are created and managed in Cloud KMS within the customer's own GCP project. Kanawai AI services reference these keys for all data encryption operations.
- Key hierarchy and wrapping — data encryption keys (DEKs) are wrapped by the customer's key encryption keys (KEKs) in Cloud KMS, following Google's multi-layer envelope encryption architecture.
- Automated key rotation — customers can configure automatic rotation schedules within Cloud KMS. When rotated, new data is encrypted with the new key version while existing data remains accessible via prior versions.
- Key revocation = data inaccessibility — if a customer disables or destroys their CMEK key, Kanawai AI immediately loses the ability to decrypt any associated data, providing a cryptographic "kill switch" for data sovereignty.
Even without CMEK, all data at rest within Kanawai AI is encrypted using Google Cloud's default encryption with AES-256 (or AES-128 at minimum). CMEK provides an additional layer of customer-controlled key management on top of this baseline.
AI Data Privacy & Protection
Kanawai AI takes an uncompromising approach to AI data privacy. Customer data ownership is absolute, and no customer data is ever used to train generative AI models.
Kanawai AI maintains a strict zero-training-data policy. No customer data — including prompts, responses, documents, or metadata — is ever used to train, fine-tune, or improve any generative AI model. This applies to all AI providers used by the platform.
Data Abstraction & Protection
- PII/PHI abstraction — before any data is transmitted to an AI model, Kanawai AI applies automated abstraction layers that detect and redact personally identifiable information (PII), personal health information (PHI), financial account data, and other sensitive data categories.
- Data rights ownership — all data rights remain strictly and exclusively with the customer. Kanawai AI processes data solely to deliver the contracted service and maintains no derived rights, licensing claims, or usage rights over customer data.
- AI model provider agreements — Kanawai AI maintains contractual data processing agreements with all AI providers (including Google Vertex AI, Anthropic, and OpenAI) that explicitly prohibit the use of API-transmitted data for model training.
- Ephemeral processing — data sent to AI models for inference is processed in memory and is not persisted in the model provider's infrastructure beyond the duration of the API request.
API & Protocol Security
All external and internal APIs within the Kanawai AI platform are secured using industry-standard protocols and hardened against common attack vectors.
Transport & Protocol Security
- TLS 1.3 enforcement — all data in transit is encrypted using TLS 1.3 (with fallback to TLS 1.2 where required). Older protocol versions (TLS 1.0, 1.1, SSL) are explicitly disabled across all endpoints.
- Mutual TLS (mTLS) available for service-to-service communication, providing bilateral authentication between Kanawai AI services and customer infrastructure.
- API authentication via OAuth 2.0 bearer tokens with short-lived access tokens (1-hour expiry) and cryptographically signed refresh tokens. API keys are supported for machine-to-machine integrations with configurable scoping and rate limits.
- Rate limiting and throttling — all API endpoints enforce per-client rate limits to prevent abuse, with automatic escalation alerts for anomalous traffic patterns.
Model Context Protocol (MCP) Security
- Sandboxed execution — MCP tool calls execute within isolated runtime environments with restricted filesystem, network, and system-call access.
- Input validation — all inputs to MCP servers are schema-validated and sanitized before execution, preventing injection attacks and unauthorized parameter manipulation.
- Audit logging — every MCP tool invocation is logged with full request/response metadata, creating an immutable audit trail for compliance and forensic analysis.
Compliance & Certifications
Kanawai AI maintains rigorous compliance with leading cybersecurity frameworks and regulatory standards, ensuring that our platform meets the requirements of the most security-conscious enterprises.
SOC 2
Type 1 — Compliant. Kanawai AI has achieved SOC 2 Type 1 certification, validating that our security controls are designed and implemented appropriately at a specific point in time.
Type 2 — In progress. We are undergoing continuous SOC 2 Type 2 compliance auditing to validate that our controls operate effectively over an extended period.
HIPAA
Kanawai AI maintains strict HIPAA data framework compliance across all cloud operating environments, ensuring that protected health information (PHI) is handled with the appropriate technical, physical, and administrative safeguards required under the HIPAA Security Rule.
NIST Cybersecurity Framework
Our security practices align with the NIST Cybersecurity Framework (CSF), implementing controls across all five core functions: Identify, Protect, Detect, Respond, and Recover. We also adhere to NIST SP 800-53 security control families for information system security.
MITRE ATT&CK
Kanawai AI integrates MITRE ATT&CK framework best practices into our threat detection and incident response processes. Our security monitoring maps adversary tactics, techniques, and procedures (TTPs) to the ATT&CK knowledge base, enabling structured threat intelligence and proactive defense.
GCP-Inherited Security
Kanawai AI is built on Google Cloud Platform, inheriting Google's world-class infrastructure security. The following controls are provided by GCP and are foundational to our security architecture.
Encryption
- Encryption at rest — all data stored in GCP is encrypted by default using AES-256 (or AES-128) with Google's multi-layer key management hierarchy. Encryption keys are themselves encrypted with a set of master keys protected by Google's robust security infrastructure.
- Encryption in transit — all data moving between GCP services and between GCP and the public internet is encrypted using TLS. Internal Google network traffic between data centers uses an additional layer of encryption (Application Layer Transport Security — ALTS).
Infrastructure & Access
- Physical security — Google operates purpose-built data centers with multi-layer physical security including biometric access controls, 24/7 monitoring, laser-beam intrusion detection, and secure server decommissioning processes.
- Identity-Aware Proxy (IAP) — access to internal management interfaces is protected by Google's IAP, enforcing identity-based access control at the application level without requiring a VPN.
- Cloud IAM — fine-grained identity and access management with support for conditions, permissions boundaries, and attribute-based access control (ABAC) across all GCP resources.
- VPC Service Controls — security perimeters that prevent data exfiltration from GCP services, enforcing that data access requests originate from authorized networks and identities.
- Security Command Center — centralized vulnerability and threat detection across all GCP resources, providing continuous security posture monitoring with automated alerting and remediation workflows.
Operational Security
- DDoS protection — Google Cloud Armor provides built-in DDoS mitigation for all internet-facing services, leveraging Google's global edge network to absorb and filter malicious traffic.
- Audit logging — Cloud Audit Logs capture admin activity, data access, system events, and policy denied events across all GCP services, providing comprehensive forensic visibility.
- Binary Authorization — ensures that only trusted, verified container images are deployed to production environments, preventing supply chain attacks at the deployment layer.