# Security

Source page: https://hrvoicenotes.tech/security/

Page content last updated October 24, 2026. This page published 2026-11-20.

**In short.** The app runs in the United States, each customer's records are kept apart from every other customer's, and access follows role. Voice recordings are deleted once the note exists. AI suggests and a person confirms. HRVoiceNotes does not hold SOC 2 or ISO 27001 certification.

HRVoiceNotes stores sensitive employee documentation, including coaching notes, incident reports and disciplinary records. This page describes, in plain terms, how that information is protected.

## Hosting and encryption

The app runs on Google Cloud and its database runs on Neon, both in the United States. Those providers encrypt data in transit (HTTPS) and at rest. Every provider we use is listed on the [Subprocessors](https://hrvoicenotes.tech/subprocessors/) page.

## Separation between organizations

Each request to the app is restricted to the organization of the person signed in. One customer's records are never visible to another customer, and automated tests verify this separation whenever the code changes.

## Access by role

- Access depends on role. Supervisors see notes about their own team. Executives and administrators see more broadly.
- A note can be marked **executive-only**, which hides it from supervisors completely.
- The full word-for-word transcript is visible only to the person who recorded it and to administrators. Other people with access see the structured note.
- New users join through one-time setup links that expire. Passwords are never sent by email.

## Deletion of voice recordings

Audio is deleted as soon as the note is generated, typically within a minute, and any residual copy is purged within 24 hours. We keep the written record, not the recording.

## AI and your data

- We use the business services of OpenAI (transcription, with Mistral AI as a backup) and Anthropic (organizing notes), not their consumer apps. None of them use the data we send to train their models, and neither do we.
- The AI suggests and a person confirms. If the AI is unsure which employee a note is about, the note is held for the supervisor to confirm.
- For disciplinary notices, the AI suggests only labels and related earlier notes. It never writes corrective actions or consequences.
- Photos attached to notes are never sent to an AI service, and location and camera details are stripped from photos before storage.
- If your organization switches on the optional Claude connector, each user who links their own Claude account can read only what they can already see in the app, and cannot change anything. The user approves each connection on a consent screen, and access ends when the user is deactivated.

## When our staff look at an account

Flourish Press staff open a customer's account only to provide support or operate the service, and every such access is logged. Note and transcript text is removed before anything reaches our monitoring tools and error logs.

## New features start switched off

Features that handle additional information, such as photos, disciplinary notices and the Claude connector, are off for every organization by default. We turn them on only when the organization asks.

## Certifications and security questionnaires

HRVoiceNotes does not currently hold a SOC 2 or ISO 27001 certification. If your organization has a security questionnaire or needs a data processing agreement, send the request through the contact form at [/apply/](https://hrvoicenotes.tech/apply/) using the topic "Privacy or data request", and we will work through it with you.

## Reporting a security problem

If you think you have found a security problem, report it through the contact form at [/apply/](https://hrvoicenotes.tech/apply/) and choose the topic "Security report". While you investigate, please do not access or change other people's data.
