> ## Documentation Index
> Fetch the complete documentation index at: https://docs.truu.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Product security

> How TruU IDV is built to be safe to connect: read-only directory access, deterministic verification decisions, tenant isolation, secrets that never surface, and an audit trail for everything.

This page describes the security properties of TruU IDV as a product. For certifications and the subprocessor list see [Compliance](https://docs.truu.ai/idv/security/compliance.md); for what data is stored and for how long see [Data handling](https://docs.truu.ai/idv/security/data-handling.md).

## Property 1: TruU reads your systems, it does not write to them

The directory connection is customer-owned and read-only. Entra ID is connected through an app registration you create with `User.Read.All`, `Group.Read.All`, and `Organization.Read.All`. TruU requests no write scope and cannot modify, create, or delete directory objects. Revoking the app in your tenant severs the connection immediately.

The same holds for connectors. ServiceNow receives events from TruU; TruU never reads ServiceNow tables. Windows TAP is the one deliberate write: when you enable it, TruU issues a Temporary Access Pass through the permission you grant for it (`UserAuthMethod-TAP.ReadWrite.All`), and nothing else.

## Property 2: Verification decisions are deterministic

Whether a person passes is decided by code, not by a model. Each factor returns a result against a configured threshold (a face match score, a code entered correctly, an answer matching a directory attribute, a document check outcome from the proofing vendor), a column passes only when every factor in it passes, and the flow passes when a column does. Thresholds, retry budgets, and lockouts are configuration you can read in the console and that are recorded on every request.

Machine learning is used inside individual factors (face comparison and liveness, document authenticity) by TruU's subprocessors, which are listed in the DPA (see [Compliance](https://docs.truu.ai/idv/security/compliance.md)). It is never used to decide which flow someone gets, to override a failed factor, or to generate a verification outcome.

## Property 3: The people who request verifications cannot configure them

Three roles, which can be combined (for example **Admin & Agent**): **Owner** and **Admin** configure the workspace (factors, flows, use cases, directory, connectors, storage, team, API clients); only an Owner can recover SSO or delete the workspace. **Agent** creates and watches verification requests for the users they are helping. See [Permissions and roles](https://docs.truu.ai/idv/admin/team-and-roles.md). Agents see the contact details on file when sending a link, the request timeline, and the outcome. They do not see directory attributes, thresholds, or the tenant-wide audit trail. The ServiceNow app runs under a machine client limited to the `idv:verification` scope, so a compromised instance can create and observe requests but cannot change a flow or a use case.

Machine clients cannot be granted administrative scopes at all.

## Property 4: The verification page cannot be steered by the caller

A verification link is minted for a specific person, use case, and flow. The user on the link cannot choose a different subject or a weaker flow; **Try another way** only offers the fallback columns the flow already allows. Links expire after a window set by the use case (see [Verification requests](https://docs.truu.ai/idv/build/verification-requests.md)), a resend refreshes the existing link rather than creating a second one, and where exclusive sessions are enabled the first browser to open a link holds it. Failed attempts count against a per-link budget and a per-person lockout.

By default, delivery goes to the contact channel already in your directory, so an attacker cannot redirect verification to a phone they control. Agents can enter a different destination; the guidance for agents is to prefer the contact on file.

## Property 5: Secrets never surface

Directory, storage, SSO, and connector secrets are held in a managed secrets service. API keys are stored hashed and shown once. No API returns a stored secret. Rotation retires the old value, with a grace window for API keys so a rollout does not break.

## Property 6: Tenant isolation everywhere

Every record, event, configuration object, and stored asset carries a tenant identifier, and every query is scoped to the requesting organization. Asset references are validated against the owning organization on every read; a reference to another tenant's object is treated as not found. Agents are further scoped to the requests they are handling.

## Property 7: You choose where evidence lives

Until you connect your own store, verification artifacts you choose to keep (proofing decision records, face references from proofing, capture media) are written to TruU managed storage partitioned per organization. You can instead connect your own Amazon S3 or Azure Blob and TruU writes there using your credentials, with optional per-object encryption under your own key. Audit records stay in TruU's database and can be exported as CSV or JSON Lines to your SIEM.

## Property 8: Everything is audited

Configuration changes, connector linking and rotation, verification lifecycle, deliveries, agent actions, and sign-ins land in an append-only, tenant-scoped audit trail with a stable event schema and OCSF mapping. PII in audit events is masked; sensitive payloads are held in controlled storage that the event references, with role-governed visibility. See [Audit events](https://docs.truu.ai/idv/security/audit-events.md).

## What TruU calls, and what calls TruU

| Direction | Hosts | Used for |
| - | - | - |
| TruU → Microsoft | `login.microsoftonline.com`, `graph.microsoft.com` | Directory reads, TAP issuance when enabled |
| TruU → ServiceNow | `*.service-now.com` (your instance) | Event push, OAuth 2.0 client credentials |
| TruU → proofing vendors | The vendor's official API host | Document and selfie checks |
| TruU → cloud infrastructure | TruU's cloud provider | Hosting, face comparison and liveness, TruU managed storage |
| TruU → your storage | Your bucket endpoint | Artifact writes when BYO storage is connected |
| TruU → messaging providers | Provider API hosts | SMS and email delivery |
| Your systems → TruU | `agent.truu.ai` (443) | API, ServiceNow app, passwordless pairing |
| End users → TruU | `verify.truu.ai`, `agent.truu.ai` (443) | The verification page and its session calls |

No inbound connection into your network is required for any integration. On-premises directories are not connected directly today.

## Human control

Agents watch requests in real time and can cancel them. Requests that exhaust their attempt budget fail; no one can override the outcome. Owners can disable any factor, flow, connector, or the directory connection at any time, and can delete a workspace's data through the domain deletion process.

> [card] [Data handling](https://docs.truu.ai/idv/security/data-handling.md)
>
> What is stored, masked, and never stored.

> [card] [Compliance](https://docs.truu.ai/idv/security/compliance.md)
>
> Certifications and third parties.

> [card] [Permissions and roles](https://docs.truu.ai/idv/admin/team-and-roles.md)
>
> The full role and scope reference.

> [card] [Threat model](https://docs.truu.ai/idv/security/threat-model.md)
>
> Attacks in scope and which factors address them.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.