HIPAA-Friendly by Architecture, Not Policy

Every time a physician dictates a note into a cloud-based tool, that audio travels to a server somewhere. It contains the patient's name. Their diagnosis. Their medications. Their history. All of it — unencrypted in transit for a moment, then processed by infrastructure the physician doesn't control, stored by a company with its own breach surface and its own retention policies.

This is the default architecture of voice dictation in medicine. And physicians largely accept it because the alternative — accurate local speech recognition — wasn't viable until recently.

It is now. That changes the calculus significantly.

The problem with “HIPAA-compliant” cloud tools

When a cloud dictation vendor says their tool is HIPAA-compliant, they mean they'll sign a Business Associate Agreement. The BAA is a contract acknowledging that they handle PHI on your behalf and accept certain obligations if something goes wrong. It doesn't prevent a breach. It assigns liability after one.

Dragon Medical One has had documented security incidents. Other cloud-based AI scribes have faced scrutiny over how long they retain audio, what they train on, and who has access to de-identified records. The BAA covers the legal exposure. It doesn't cover the patient whose information was compromised.

Resonant doesn't require a BAA because there's nothing to sign a BAA about. Your audio never reaches a third party. Processing happens on the Neural Engine inside your Apple Silicon Mac, the audio is discarded immediately after transcription, and the text appears in your EHR. No transmission. No retention. No exposure surface.

HIPAA-friendly by architecture means the compliance question doesn't arise — not that it's been addressed and managed.

Two hours a day is a career-level problem

Physicians spend more than two hours on documentation for every hour of direct patient contact. That number has climbed as EHRs have grown more complex and more mandatory. The term “pajama time” — charting after the kids are in bed — has its own entry in the burnout literature.

Voice dictation addresses the right bottleneck. Most of that documentation time isn't thinking — it's translating thought into typed text. Speaking runs three to four times faster than typing for most people. An HPI that takes five minutes to type takes ninety seconds to dictate.

The compounding effect across a career is substantial. A physician who dictates efficiently for twenty years recaptures roughly a month of working hours per year. That's not a productivity metric. That's time that goes back to patients, or to rest.

Hospital IT is not your ally

Many hospitals and health systems block or restrict outbound traffic from clinical workstations. Cloud-based dictation tools require internet access to function — audio has to reach the server. In environments where that traffic is throttled, filtered, or blocked entirely, the tool either fails silently or introduces unacceptable latency.

Resonant requires no outbound connection. Everything runs on the machine. That means it works in the OR, in an electronically isolated exam room, in a basement clinic, or in a rural facility with unreliable connectivity. If your Mac is on, Resonant works.

It also means no IT approval process for a cloud vendor. No waiting for the security review. No BAA negotiation with your compliance department. Install it and use it.

What this looks like in practice

The workflow is one hotkey. Click into the target field in your EHR — the HPI section, the assessment and plan, a referral letter — trigger Resonant, speak, trigger again to stop. Text appears where your cursor is. No app switching, no dedicated dictation interface, no cloud round-trip latency.

Accuracy on medical terminology is strong out of the box. Drug names, procedures, anatomical terms, and diagnostic language are handled well by the underlying models. You'll correct the occasional word, as with any transcription tool. The net speed is still significantly faster than typing.

Physicians who adopt dictation consistently report that notes end up better organized and more detailed, not just faster. Speaking a note the way you'd explain a case to a colleague — conversational, structured, direct — produces better documentation than translating the same information into typed prose under time pressure.

Why local processing is the right default for medicine

The cloud became the default in voice dictation because local speech recognition used to be slow and inaccurate. That constraint is gone. Apple Silicon's Neural Engine runs modern Whisper-based models in real time without a network connection. The hardware that ships in a MacBook Pro today is capable of the same transcription quality that required cloud infrastructure three years ago.

Medicine should have a higher bar than most fields for how patient data is handled. A tool that eliminates the data exposure entirely — not through policy, but through architecture — is the better default. The BAA requirement, the security review, the breach risk, the questions about training data and retention: none of these arise if the audio never leaves the machine.

Resonant is a free download for Mac. No account required. No subscription. Works offline. Your notes stay yours.

Download Resonant and dictate your first note today.