Skip to main content
Founder-Led · Practicing Since 1997 Founder-led. Specialist-delivered. Every engagement follows standards developed by Tony Paris over 29 years.
Voice, Chat & Business Intelligence

Build secure two-pass verification

By Tony Paris · November 5, 2026 · 2 min read

Drafted with AI and published by AppWT.

I explain a simple two pass pattern that keeps risky actions under human control. You can use it for voice agents, chatbots, and internal tools without sending data off network.

I use two pass verification to keep automation safe. The system proposes an action, then waits for a one time code and a human approval before it runs. This pattern works for voice, chat, and internal tools, and it keeps your data on your network. You can add it to high risk moves like refunds, exports, payouts, and account changes with a clear plan.

Plan the flow

  • List actions that always need human approval, such as issuing credits, changing inventory, editing user roles, or exporting files.
  • Choose channels you control for codes and approvals, such as company email, an internal chat space, or an in portal approval screen.
  • Define who can request and who can approve. Separate requester and approver when possible and document backups.
  • Decide what context to show approvers, like requestor, action, target, limits, and a short summary.

Generate and deliver the code

  • Use a cryptographically secure random generator to create a short code. A one time code is a secret that works only once and then expires.
  • Hash the code with a one way function such as SHA 256 and store only the hash with requestor, action details, and a brief expiration time.
  • Send the code on a controlled channel along with the action summary and a unique request ID. Keep the full data on your network.

Approve and execute

  • Require the approver to read the summary and enter the code or reply Approve plus the code. Verify the code, request ID, and that details match and are not expired.
  • Rate limit attempts, then freeze the request if guesses continue. Log requester, approver, timestamps, action parameters, and outcome.
  • Execute only the approved step. Use allow lists and fixed runbooks so the agent cannot add extra commands.

Voice and chat patterns

  • For phone, an AI receptionist can confirm the action, collect the code from the approver, and repeat the summary before running it.
  • For chat, post the pending request in an internal channel and wait for Approve plus code from a known approver account.

What to check

  • Wrong code, expired code, reused code, or mismatched details all fail safely and leave a clear audit trail.
  • No code or action data leaves controlled channels. Logs are complete and stored on your network.
  • Approver separation holds, attempts are limited, and you can review and replay the audit path in tests.

If you want help, I design and implement this pattern using Voice and Chat Full Conversational AI Agent, Voice and Chat AI Phone, AI and Automation AI Agents and Personas, and AI and Automation Business Automation, including AI phone and chat agents that fit your environment.

Get the next one in your inbox

Pick your topics and how often, from weekly to quarterly. I send a confirmation email first, and you can change or stop it any time.

Get more like this by email

Bank deposit reconciliation win

I show how I automated bank deposit reconciliation on the client network with a simple chat agent and clear matching rules.

All Voice, Chat & Business Intelligence articles

Share This Page
Awards & Recognition
AI-generated AppWT theme song art: two singers in black and gold on a rooftop above the Detroit skyline at sunset

Accessibility

by AppWT Web & AI Solutions
🛡️ Accessibility Profiles
📝 Content Adjustments
100%
100%
1.4
0px
🎨 Color Adjustments
100%
🎛️ Orientation & Controls
♿

Accessibility Statement

Our commitment to digital accessibility and inclusive design

Our Commitment to Accessibility

AppWT Web & AI Solutions is committed to ensuring digital accessibility for people with disabilities. We continually improve the user experience for everyone and apply the relevant accessibility standards to achieve these goals.

Conformance Status

The Web Content Accessibility Guidelines (WCAG) defines requirements for designers and developers to improve accessibility for people with disabilities. It defines three levels of conformance: Level A, Level AA, and Level AAA.

AppWT Web & AI Solutions is partially conformant with WCAG 2.1 level AA. Partially conformant means that some parts of the content do not fully conform to the accessibility standard.

Accessibility Features

  • Built-in accessibility toolbar with multiple customization options
  • Keyboard navigation support throughout the website
  • Screen reader compatibility and proper ARIA labels
  • High contrast mode and color customization options
  • Text size adjustment and font modification capabilities
  • Reading guide and focus indicators for improved navigation
  • Alternative text for all images and media
  • Semantic HTML structure for better screen reader interpretation

Technical Specifications

Accessibility of AppWT Web & AI Solutions relies on the following technologies to work with the particular combination of web browser and any assistive technologies or plugins installed on your computer:

  • HTML
  • WAI-ARIA
  • CSS
  • JavaScript

These technologies are relied upon for conformance with the accessibility standards used.

Feedback

We welcome your feedback on the accessibility of AppWT Web & AI Solutions. Please let us know if you encounter accessibility barriers:

Phone: (888) 565-0171

Email: sales@appwt.com

Address: 33300 Five Mile Rd, Livonia, MI 48154 (by Appointment Only)

Assessment Approach

AppWT Web & AI Solutions assessed the accessibility of our website by the following approaches:

  • Self-evaluation
  • External evaluation
  • Automated testing tools
  • Manual testing with assistive technologies

Date

This statement was created on January 15, 2025 using the W3C Accessibility Statement Generator Tool.

Last updated: