> ## 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.

# Threat model

> The attacks TruU IDV is designed to stop at the help desk and during onboarding, which control addresses each one, and where the honest limits are: deepfakes, SIM swap, stolen IDs, phished links, insiders with directory access.

This page is for the security reviewer deciding whether TruU IDV covers the attacks that worry them. When you finish, you can map each attack to the control that addresses it and say out loud what the product does not claim.

## The controls, named once

| Control | What it is |
| - | - |
| **Trusted reference photo** | The Face factor compares the live person against a photo TruU did not get from the caller: your directory photo, a photo in your storage, or the selfie from a passed document check. |
| **Liveness** | Face checks stream video to AWS Rekognition for liveness before comparison. Document checks run the vendor's own liveness on the selfie. |
| **Name binding to the directory** | A document check passes only if the name on the document matches the directory record the request was created for. |
| **Delivery to directory contacts** | By default, verification links and one-time codes go to the email or phone already on the user's directory record. The caller cannot supply an address. An agent can enter a different destination when needed; train agents to prefer the contact on file. |
| **Link lifetime and single use** | Help desk links expire in 30 minutes, reset links in 60, onboarding links in 7 days. A link completes once; a resend refreshes the same request rather than minting a second one. |
| **Attempt limits** | Each factor and each request has a limited number of attempts. Exhausting them fails the request; it does not reset. Repeated failures lock the person out of new sessions for a period. |
| **Audit** | Every search, preview, request, factor attempt, fallback, and cancel action is an append-only event with actor, user, target, and outcome. |

## Attacks in scope

### Help desk social engineering

**Attack.** The caller talks an agent into a password reset or MFA re-enrollment with a plausible story, personal details scraped from LinkedIn, and urgency.

**Controls.** The agent sends a verification instead of judging the story. By default the link goes to the directory contact, so the caller must hold the real user's phone or mailbox. The use case's rules pick a flow by directory group, so an executive gets Face plus Proofing regardless of how convincing the call is. Audit records the agent's search, preview, and request so the sequence is reconstructable. Overrides are not a button; a failed request stays failed.

**Limits.** Knowledge questions drawn from directory attributes are weak against a researched attacker. Use them only as a fallback and only on attributes that are not public.

### SIM swap

**Attack.** The attacker ports the user's number and receives SMS codes and links.

**Controls.** SMS is one factor among several, and a column that contains Face or Proofing cannot be passed by SMS alone. Use case rules can keep SMS out of the flows high-value groups get. Email delivery to the directory mailbox is unaffected by a port.

**Limits.** SMS is weaker than email where phone numbers are poorly maintained, and in most directories they are. If a column is SMS alone, a SIM swap defeats it. Treat SMS as a convenience factor, not a security one.

### Deepfake video and camera injection

**Attack.** A synthetic face, a replayed recording, or a virtual camera feeding the browser.

**Controls.** Liveness runs on streamed video, not a single uploaded frame, and the comparison is against a trusted reference the attacker did not supply. Document checks run the vendor's own injection and presentation-attack detection on the selfie.

**Limits.** No biometric method is perfect. Liveness vendors publish detection rates, not guarantees, and injection techniques improve. Pair Face with a second factor in the same column for the populations that matter most, and keep attempt limits tight so an attacker cannot iterate.

### Replayed selfies

**Attack.** A photo or video of the real user, taken from social media, shown to the camera.

**Controls.** Liveness is designed for exactly this: a printed photo or a screen replay fails the depth and motion checks. The match score is only consulted after liveness passes.

**Limits.** The same as above. Liveness reduces the attack to the hard case; it does not remove it.

### Stolen or borrowed ID document

**Attack.** The attacker presents a real document that is not theirs, or a document bought to match a target name.

**Controls.** The Proofing factor requires a selfie that matches the document photo, with liveness. Name binding rejects a genuine document in the wrong name. When **Proofing photo** is on, the selfie from a passed check becomes the face reference, so a later Face check compares against the person who held the document, not a stale directory photo.

**Limits.** A stolen document plus a lookalike attacker is the residual case. Document checks are as good as the vendor's authenticity analysis for that document type and country.

### Phishing of verification links

**Attack.** The attacker sends a fake verification message to harvest credentials, or convinces the user to forward a real link.

**Controls.** A verification link never asks for a password. It is single use, short lived, and bound to a specific request and use case. The verification page is branded under your Branding settings so a lookalike is easier to spot. Where exclusive sessions are enabled for your workspace, the first browser to open a link holds it.

**Limits.** TruU cannot stop a user from acting on a message that did not come from TruU. Brief users that a verification link arrives only after they, or an agent they are talking to, requested one. See [What to expect during verification](https://docs.truu.ai/idv/end-users/what-to-expect.md).

### Insider with directory write access

**Attack.** Someone who can edit Entra ID changes a target's phone, email, or photo, then requests a verification that routes to themselves.

**Controls.** Directory changes are outside TruU, but the request is not. Audit shows who requested the verification and which contact value it went to, masked, so a reviewer can line it up against the directory change log. A trusted reference photo from a passed document check does not change when the directory photo does. Face and Proofing columns do not depend on contact values at all.

**Limits.** TruU reads your directory and trusts it. An attacker who controls the directory can weaken delivery-based factors. Put Face or Proofing in the primary column for privileged populations, and alert on contact-attribute changes in your own directory audit.

### Link forwarding

**Attack.** The real user receives a legitimate link and hands it to someone else, willingly or under pressure.

**Controls.** Face and Proofing bind the link to a person, not a device: the forwarded link still has to pass liveness against the user's reference. Short lifetimes shrink the window. Where exclusive sessions are enabled, a second browser is refused while the first holds the session.

**Limits.** A user who sits next to the attacker and completes the face step for them has not been impersonated; they have colluded. That is an HR and conduct problem, and the audit trail will show the user passed.

## What TruU IDV does not claim

* **No biometric method is perfect.** Liveness and face comparison have false accept and false reject rates. TruU shows factor strength and assurance bands, not a guarantee.
* **Knowledge questions are weak.** They depend on directory data the attacker may have researched and that is often stale. Use them as a fallback for low-risk populations.
* **SMS is weaker than email** where numbers are poorly maintained, and it is exposed to SIM swap. Check coverage before building a column on it.
* **TruU trusts your directory.** A compromised directory weakens any factor that reads from it.
* **TruU does not detect compromise of the user's own device.** A malware-controlled phone that passes a live face check is the user, as far as the check can tell.

## Choosing columns against this model

| Population | Primary column (A) | Fallback (B) | Why |
| - | - | - | - |
| Privileged admins, finance, executives | Face + Email | Proofing | Two independent bindings; no SMS or Knowledge path |
| General workforce | Face | Email + Knowledge | Face for most; a delivery-plus-knowledge path for users without a reference photo |
| New hires with no directory photo yet | Proofing | None | With Proofing photo on, the document check creates the reference photo for later Face checks |
| Populations with poor phone data | Email + Knowledge | Proofing | Avoid SMS until numbers are cleaned up |

See [Flows](https://docs.truu.ai/idv/build/flows.md) for the column model and [How use cases work](https://docs.truu.ai/idv/use-cases/overview.md) for assigning flows to groups.

## Related pages

> [card] [Product security](https://docs.truu.ai/idv/security/overview.md)
>
> The properties that make these controls hold.

> [card] [Factors](https://docs.truu.ai/idv/factors/overview.md)
>
> Each factor's strength and requirements.

> [card] [Audit events](https://docs.truu.ai/idv/security/audit-events.md)
>
> Reconstructing an attack after the fact.


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