Skip to main content

Security

A recording archive is a high-trust thing to run. This page describes how it is protected, including the parts that are limitations rather than guarantees.

Last updated

Principles we build to

  • Recordings are the most sensitive thing in the system, so they get the strictest handling in the system.
  • Access to customer content is the exception, requires a reason, and is logged.
  • Your content is never used to train models — ours or a provider's.
  • You can always get your data out, and you can always delete it.
  • We would rather tell you a limitation than imply a guarantee we cannot keep.

Encryption

All traffic between your device and our infrastructure uses TLS, with strict security headers on every response.

Each recording is encrypted with AES-256-GCM using a key generated just for that file, with your account bound into the encryption so the file can only be decrypted for you. That key is itself encrypted with a master key held in a separate, restricted secret store, never next to the recordings or in the database. Storage only ever holds encrypted files. Transcripts and notes are encrypted at rest by our database provider.

Audio is decrypted only in memory, when you play it or while it is being transcribed. Transcription and AI requests go to OpenAI's API over encrypted connections.

Access control

Internal access to production systems follows least privilege, is limited to the people who need it to operate the service, and is protected by strong authentication.

Every request is checked against your sign-in, so a recording can only be listed, played or downloaded by its owner. Uploads, playback, data exports, deletions and Apple Watch linking are recorded in a security log, which stores a one-way hash of the IP address rather than the address itself and is kept for one year.

Access to customer content by our team requires a documented reason, normally a support request you started. Nobody browses recordings.

Subprocessors

Providing the service requires third parties for hosting, storage, speech recognition and language model inference. Each is covered by a data processing agreement that restricts them to processing your content on our instructions and prohibits training on it.

The current subprocessor list is available on request, and we notify account holders before adding a subprocessor that processes recording content.

Retention and deletion

Recordings and transcripts persist until you delete them or close your account, because long-term recall is the point of the product. You choose what to keep.

Deleting a recording removes the audio, transcript, summary and search index entries derived from it. Encrypted backups cycle on a rolling basis, so deleted content can survive briefly in backups before being overwritten. Closing your account deletes your content and reduces account data to what we must retain for legal and financial records.

Availability and backups

Data is stored on managed infrastructure with redundancy and encrypted backups. Recordings made without connectivity are queued locally and uploaded when the connection returns, so a bad network delays processing rather than losing a recording.

We do not promise uninterrupted service. If something irreplaceable is recorded, keep your own copy as well; that is sound practice with any software.

What is on your side

  • Protect your account credentials and use strong, unique authentication.
  • Obtain whatever recording consent applies where you and the other participants are, and honour requests to stop.
  • Think before recording highly sensitive material — legal, medical or compensation discussions — and know the retention model before you do.
  • Verify names, numbers and dates in a transcript against the audio before relying on them. Transcripts read fluently even when they are wrong.

Reporting a vulnerability

If you believe you have found a security issue, email zubair@akstuudio.com with the subject line "Security disclosure" before disclosing it publicly. Include the steps to reproduce and, if relevant, the affected endpoint.

We confirm receipt within two business days, keep you informed while we fix it, and are happy to credit you. We will not pursue legal action against good-faith research that avoids privacy violations, data destruction and service degradation.

Security questions we get asked

Is my audio encrypted?

Yes. Every recording is encrypted with AES-256 using its own key, bound to your account, before it is stored. Everything travels over TLS. Transcription goes through OpenAI's API, which does not train on API data and deletes it within 30 days.

Can myvoicenotes.ai staff listen to my recordings?

Not as a matter of course. Access to customer content is restricted, requires a specific reason such as a support investigation you asked for, and is logged. We do not review recordings for product development.

Are my recordings used to train AI models?

No. Your recordings, transcripts and summaries are never used to train our models or our providers' models. Provider agreements prohibit it, and it is the commitment we would least like to break.

How do I report a security vulnerability?

Email zubair@akstuudio.com with the subject line "Security disclosure" before disclosing publicly. We confirm receipt within two business days, keep you updated while we fix it, and will credit you if you would like us to.

Can I get a data processing agreement for my organisation?

Yes. Email us and we will send our DPA along with the current subprocessor list, so your procurement or legal team can review the actual chain of processing rather than a summary of it.

See also the privacy policy for what we collect and why, and the terms of service for the agreement itself.