Security & responsible AI · Review draft

Useful intelligence needs clear boundaries.

This page describes the controls that must be substantiated for public claims. It is not a certification, security audit, or contractual guarantee.

Demo isolation

The live voice demo must use a separate assistant, fictional Jackson Apartments knowledge, no customer tools or external actions, and no production phone routing. Reuse verified voice settings only; do not copy customer prompts, recordings, knowledge stores, or webhook credentials.

Approved knowledge and uncertainty

The assistant should answer from approved information, distinguish requests from confirmations, and clarify ambiguity instead of inventing availability or outcomes.

Human oversight

Dispatch, financial actions, legal decisions, and changes to live routing require explicit authority. The demo must not execute them.

Evidence needed before publication

• Account boundaries and role-access tests.
• Encryption scope, credential handling, and approved subprocessors.
• Recording and transcript policy, retention periods, deletion process, and provider-copy behavior.
• Incident contact, response process, and backup/restore evidence.
• Verified operating limits, failure paths, and human handoff tests.

Preview behavior

No analytics, advertising cookies, external fonts, or remote pilot-form submissions are used by page code. Microphone audio is sent to Telnyx only for an explicitly started voice demo; see the privacy notice. Local downloads remain under the tester’s control.