Trust & Security

Trust Center

Everything you need to approve IdentityOps during a security review. Built for CISOs, procurement, and vendor risk teams.

Security Posture Summary

Tenant access:Read-only delegated

Data stored:Metadata only

AI model:None (deterministic templates)

Tenant isolation:Database-level scoping

Offboarding:30-day restore, then permanent purge

Vulnerability reporting:security@identityops.dev

Guarantees & Boundaries

What IdentityOps commits to, and what it is architecturally unable to do.

We guarantee

Read-only Microsoft Graph permissions — zero write scopes

Delegated access only — runs as the signed-in admin, not a background service

No passwords, secrets, certificates, or tokens from your tenant are ever stored

All AI is deterministic template assembly — no LLM, no external API calls

Tenant data is isolated at the database level with foreign key scoping

Consent is revocable instantly from the Entra Admin Center

We cannot

Cannot modify, create, or delete any object in your tenant

Cannot access other tenants' data — no API path accepts a tenant ID as input

Cannot make autonomous decisions — AI explains risk, it does not enforce policy

Cannot retain data beyond 30 days after cancellation

Cannot send your data to external services, advertisers, or training pipelines

Cannot bypass Microsoft Graph permission enforcement — even with a bug

Threat Model

How IdentityOps mitigates the risks your security team evaluates.

Credential compromise

IdentityOps never stores tenant credentials. Access tokens are encrypted at rest (AES-256-GCM) and never leave the server. Sessions rotate every 15 minutes with a 24-hour absolute max.

Token leakage

Only the session ID is stored in a cookie (HttpOnly, Secure, SameSite). Refresh tokens are server-side only, encrypted, and scoped to the specific tenant. No tokens are exposed to client-side code.

Insider threat

Internal admin panel uses a completely separate authentication system. All admin actions are logged in an immutable audit trail. Customer data access is scoped per-tenant with no override path.

Infrastructure breach

Hosted on Vercel (edge runtime) with data in Neon PostgreSQL. TLS 1.2+ for all communication. No self-hosted infrastructure. All child records cascade-delete with the tenant — no orphaned data.

Supply chain / dependency

No external AI services. No third-party analytics or tracking SDKs processing tenant data. Stripe handles billing only — no tenant metadata is shared with payment processors.

Security Architecture at a Glance

Five pillars. Each answers a question your security team will ask.

Access Control

“What permissions do you need?”

Three read-only delegated Microsoft Graph permissions. No background service, no agents, no scripts installed in your environment. Even if there were a bug, Microsoft would block any write operation at the permission level.

Data Protection

“What data do you store?”

App metadata, credential expiry dates, and owner info. Never stores passwords, secrets, or certificates. Never sold, shared, or used for training.

Tenant Boundaries

“Can one tenant see another?”

Every query scoped to the authenticated tenant. Per-tenant roles. No API path accepts a tenant ID as user input. All child records cascade-delete with the tenant.

Exit & Deletion

“What happens when we leave?”

Cancel in-app or remove the Enterprise Application from Entra. 30-day self-service restore window, then automatic permanent deletion. No residual data retained.

AI Safety

“Is AI doing anything unsafe?”

All AI uses deterministic templates — no external LLM calls. Every sentence traces to a computed field. AI explains risk; it never decides or acts.