Security & Trust

Built for regulated industries
from the first line of code.

Xelurel AI handles sensitive AI outputs across healthcare, legal, and financial services. Security and auditability aren't features we added later — they're the architecture.

✓ liveData encrypted at rest
✓ liveData encrypted in transit
✓ liveAudit trail — immutable
in progressSOC 2 Type II
plannedHIPAA BAA
Data

What we store — and what we don't.

Xelurel AI never stores your raw AI outputs or user prompts. Content is hashed (HMAC-SHA256, tenant-scoped key) for integrity verification only. The governance decision, risk score, and triggered rules are stored — not the content itself.

🔒

Content hashing, not content storage

Every prompt and output is hashed using a tenant-scoped HMAC-SHA256 key before the assessment is persisted. The hash lets you verify integrity later — but the original content cannot be reconstructed from it.

📋

What is stored per decision

Decision ID, tenant ID, decision outcome (allow / review / block), risk score (0–100), triggered rule IDs, policy ID and version, API key environment, and timestamps. No raw text.

🗑️

Data retention

Default retention is 12 months. Enterprise customers can configure custom retention windows under a data processing agreement. Deletion requests are honoured within 30 days.

🌍

Data residency

All data is stored in Google Cloud Firestore (us-central1 by default). Enterprise customers on custom contracts can request regional isolation.

Encryption

Encrypted everywhere, always.

At rest

  • AES-256 encryption via Google Cloud Firestore
  • Tenant-scoped HMAC-SHA256 keys for content hashing
  • API keys stored as SHA-256 hashes — never in plaintext
  • Webhook signing secrets stored encrypted

In transit

  • TLS 1.2+ enforced on all endpoints
  • HTTPS-only — HTTP requests redirected
  • HMAC-SHA256 signatures on all outbound webhooks
  • Stripe-signed webhooks for billing events
Access Control

Least privilege, enforced server-side.

👥

Role-based access control (RBAC)

Every team member is assigned a role — admin, reviewer, or read_only. Roles are enforced server-side on every API route, not just the UI. A reviewer cannot publish policies. A read_only member cannot invite teammates or call mutating endpoints.

🔑

API key security

API keys are shown once at creation, then stored as SHA-256 hashes. Keys are scoped to a single tenant and environment (test / live). You can revoke any key instantly from the dashboard. Rate limits are enforced at three independent layers: per-tenant, per-key, and per-IP.

🍪

Session management

Dashboard sessions use Firebase-issued HttpOnly, Secure, SameSite=Lax cookies with a 7-day TTL. Sessions are verified server-side on every authenticated request using Firebase Admin SDK — no client-side trust.

🏢

Tenant isolation

Every Firestore read and write is scoped to the authenticated tenant's ID. There are no cross-tenant queries. Multi-tenant admin access is available to super-admins only, identified by allow-listed email at session creation.

Audit Trail

Every decision. Immutable. Exportable.

Xelurel AI's core function is producing an auditable record of every AI governance decision. That record is designed to satisfy compliance reviewers, not just developers.

Decision ID

UUID v4 — unique per assessment

Timestamp

ISO 8601, server-side (not client clock)

Decision

allow / review / block — immutable once written

Risk score

0–100 integer + normalized float

Triggered rules

Rule IDs + human-readable reasons

Policy version

Exact version applied at assessment time

API key env

test or live — prevents test/prod confusion

Content hashes

HMAC-SHA256 of prompt + output

Reviewer actions

Override decision + note, timestamped

Decision logs can be exported as CSV or JSON directly from the dashboard, with optional date range filtering. Enterprise customers get audit exports formatted for regulatory submission.

Compliance

Where we are and where we're going.

SOC 2 Type IIIn progress

Audit in progress. Controls are designed to Type II standards. Report available to enterprise customers under NDA upon request.

HIPAA BAAAvailable on Enterprise

Available to Enterprise customers handling PHI. Contact us to execute a Business Associate Agreement. Healthcare policy templates are available on all plans.

GDPRCompliant

Xelurel AI does not store raw personal data. Hashed content cannot be reversed. Data deletion requests honoured within 30 days. DPA available on request.

Audit log retentionLive

Default 12-month retention on all paid plans. Custom retention windows (longer or shorter) available under Enterprise data processing agreements.

Pen testingScheduled

Annual third-party penetration testing. Results available to enterprise customers under NDA.

Infrastructure

How we run it.

Hosting

Vercel Edge Network

Global CDN · automatic TLS · zero cold-start for API routes

Database

Google Cloud Firestore

Multi-region replication · automatic backups · point-in-time recovery

Auth

Firebase Authentication

Industry-standard token flow · short-lived ID tokens · HttpOnly session cookies

Rate limiting

Upstash Redis

Sliding window · cross-instance · 3 independent layers per API request

Email

Resend

Transactional only · no marketing lists · SPF + DKIM + DMARC configured

Payments

Stripe

PCI-DSS compliant · card data never touches Xelurel AI servers

Disclosure

Found something? Tell us first.

We take security reports seriously and respond to all valid reports within 5 business days. We ask that you give us reasonable time to investigate and remediate before public disclosure.

Responsible disclosure

Report security vulnerabilities to security@xelurel.com. Include steps to reproduce, impact assessment, and your contact details. We do not currently offer a bug bounty program but acknowledge all valid reports.

Questions about our security posture?

Enterprise security reviews, custom DPAs, and BAA negotiations handled directly. We share our controls documentation, architecture diagrams, and pen test summaries under NDA.

Contact security team →View Enterprise plan