How AI is controlled
AI Governance
How the AI features in SecurePoint USA are controlled today.
What is in place
People decide
AI output never clears or holds a visitor. A held visit is cleared or denied by an authorized person.
Audited
Decisions are written to an append-only audit trail.
Model recorded
Each AI summary and each ID read is stored with the provider and model that produced it.
Off switches
An administrator can turn the ID read off for their organization, and a server setting turns it off for every organization.
Change control
- AI prompts and model settings live in the code
- Code changes reach production only through pull requests with required checks
When AI is unavailable
- Screening completes without a summary
- The ID read can be turned off for every organization at once
Records
Each AI summary is stored on its screening record with the provider and model version, and each ID read is audited with the model that read it. The reviewer's decision and reason are in the append-only audit trail.