Skip to content
Transparency & Trust

AI Model Cards

What each AI feature in SecurePoint USA does, what it reads, and who decides. Name matching is listed too, because it is not AI.

Name Matching (not AI)

Intended Use

Find possible matches between a visitor and the screening lists. This is deterministic fuzzy matching, not an AI model

Inputs

Visitor name and the screening list records

Outputs

Match candidates with a similarity score, not determinations

Oversight

A held visit is cleared or denied by an authorized person

Limitations

Ambiguous names; transliteration variance

Risk Summary Assistant

Intended Use

Summarize each screening for the reviewer

Inputs

The screened name, company and country

Outputs

A short summary stored with the provider and model version

Oversight

Advisory only; it never clears or holds a visitor

Limitations

May omit details; relies on source quality

ID Text Reader

Intended Use

Read the text on the front of a visitor's ID at a browser kiosk to fill in their details. The SecurePoint Kiosk apps never use it

Inputs

A photo of the front of the ID. The visitor is not asked first

Outputs

Parsed text fields that the visitor checks

Oversight

Runs for every organization until an administrator turns it off; the visitor checks every value it fills

Limitations

Image quality, glare, occlusion

These model cards are living documents, updated as our systems evolve. Last reviewed: November 2026