AI trust and data

Your data trains nothing.

Before you let AI near a session, you should know where your clients' data goes. This page answers that in plain English, with five promises and a note on what binds us to each one.

A 20-minute demo with no commitment. Read the No AI Therapist Pledge

Most therapists have already made up their minds about software for the easy parts of admin, like reminders and superbill formatting. Clinical work is different, and the reason has little to do with saving time. The question there is whether you could explain the tool, and defend it, to the client sitting across from you.

An EHR is the hardest place to answer that honestly, because it holds the most sensitive material in your practice: the session itself, in your words and your client's. The bar for trust is higher here than anywhere else in your software. So this page starts with where your clients' data goes and where it never goes, and leaves what our AI can do for the feature pages.

One habit is common across the industry. When a vendor reassures you about AI and data, watch the adjective. The promise is rarely "we won't use your clients' data." It is usually "we won't use your clients' identifiable data," and that one word leaves room for everything else. The five promises below don't rely on it. Each one covers only what we control, and you can hold us to every one.

Promise 1

We sign a BAA

The worry it answers: whether we are legally on the hook for your clients' data.

A Business Associate Agreement is the contract HIPAA requires whenever a vendor creates, receives, maintains, or transmits protected health information on your behalf (45 CFR Part 164, and the HHS HIPAA for Professionals guidance). It makes a vendor legally accountable for the data, with real liability behind it. CoralEHR signs one. The practice owner signs it inside CoralEHR, and the current text is published at /baa/. For why a BAA matters in solo practice, our HIPAA guide for private-practice therapists walks through it. The AI provider that powers our note drafting operates under that same BAA chain of subprocessors.

A BAA is the minimum the law requires. It makes us legally liable for mishandling your data. On its own it does not promise that we won't train on your data or that we won't ask you to record sessions. Plenty of vendors sign a BAA and still keep broad data rights elsewhere in their terms. So the BAA is where this page starts, and the next four promises cover what we do beyond it.

Promise 2

No recordings without written consent

The worry it answers: whether the AI keeps a recording of your clients.

Recording a therapy session changes the room. Once a client knows their words are being saved, something shifts, and you are the one who manages that shift in every session. So the first question to ask about any AI note tool is what it keeps from the session after the draft is written. How good the draft is comes second.

When it is on, CoralEHR's notes assistant listens during a video session with one client and transcribes it to draft your note, and it never stores the audio. The transcript is deleted when you sign the note, or after 7 days if you don't. Clients agree in the portal before the assistant listens, and they can withdraw.

Sessions are recorded in CoralEHR for one reason only: supervision, and only with the client's written consent. Those recordings stay inside CoralEHR, can't be downloaded, and are deleted on a schedule, or right away if the client withdraws consent.

For contrast, from the competitor's own page: SimplePractice's Note Taker also drafts from a session transcript, and from June 2026 it keeps de-identified transcripts by default for new users, with no practice-wide switch to turn that off (transcript-retention FAQs). CoralEHR deletes the transcript once the note is signed. Keeping transcripts can be a reasonable choice for some practices, and it should be one you make on purpose.

Promise 3, the core one

Your data trains nothing

The worry it answers: whether your clients' words end up inside someone's model.

Here it is with no adjective hiding in it. Your clients' data, including notes, transcripts, assessments, messages and any other PHI, is never used to train, fine-tune or evaluate any shared or foundation model. That covers the identifiable version, any de-identified version and any aggregated version. These are Pledge planks #1 and #2, said plainly.

One condition, stated plainly. Plank #2 is a promise about how we work today and how any change would happen. Today we don't use your data this way. We won't start without your explicit, opt-in consent, which you can withdraw, and we will tell you clearly and in advance before anything changes. We won't present that as a forever promise we can't honestly make. Right now your data trains nothing, and the only way that could ever change is with your own yes.

Where your data actually goes. Our AI runs on Anthropic's first-party Claude API through one fixed provider path, so your data is never routed to other model providers, and under our agreement your inputs are not used to train the model. Speech to text runs through Deepgram. Your note content lives in the client's chart under our BAA, and it isn't sold, shared for advertising or fed to a foundation model. Assistant chats are kept for 30 days and then deleted.

Why "trains nothing" is stronger than "trains only on de-identified data." The industry's comfortable phrase is "de-identified." A psychotherapy note does not de-identify the way a billing code does. The story of a person's life, in their clinician's words, works like a fingerprint, and with the name removed the session is still recognizably theirs. "Trains nothing" leaves no such gap.

The competitor contrast, dated, with each quote linked to the page it comes from. SimplePractice's own de-identification page (published June 3, 2026) states it "may de-identify data to help build and improve tools for our customer base" and that "de-identified transcripts will be used to improve Note Taker and related AI features" under HIPAA Safe Harbor. Separately, on its transcript-retention FAQ, SimplePractice states that retained transcripts are not sold and that personal health information is never sold. Both are SimplePractice's own statements, and we represent them as written. We never say "they sell your data." The real distinction is narrower. SimplePractice reserves the right to use de-identified transcripts to improve its AI features. CoralEHR's promise is that your data is used only for your own practice, and we don't de-identify it to improve our tools either.

Promise 4

A clinician signs every note

The worry it answers: whether the AI makes clinical calls.

It doesn't. This section states only what the product does today.

  • ✓ The clinician is the author of record. Our AI drafts, transcribes and summarizes. It never diagnoses or decides. An AI-drafted note stays a preliminary draft and becomes part of the signed record only after a licensed clinician has read and signed it. (Pledge plank #4.)
  • ✓ The guardrails are written into every AI feature. The AI's instructions bar it from diagnosing, rating risk or recommending treatment, and from stating any fact that isn't in the record or the session. When it reports a feeling, it uses the words someone said or wrote down.
  • ✓ Measures are scored by fixed math. Scores for measures like the PHQ-9 and GAD-7 are calculated by fixed rules with no AI involved, and an answer on the PHQ-9 item about self-harm raises a safety alert by a fixed rule. The model has no say in either.
  • ✓ Coming features follow the same rule. Treatment-plan suggestions are coming to Pro, and each item will be a suggestion for you to accept, edit or reject before anything is saved.

The result is a human name on every clinical word. You can review everything the AI produces, and the licensed clinician carries the judgment and the accountability. In practice, care you can defend means you can say I read it, I edited it, I signed it, and my license is on it.

Promise 5

Where these promises live

The worry it answers: marketing pages can be edited, so what actually binds us.

That's fair. A web page is not a contract, and we won't pretend it is. Here is what is in writing and what is policy.

  • ✓ Your BAA is a contract. The Business Associate Agreement you sign during onboarding governs how protected health information is used and protected, with legal consequences if it is broken. Your subscription is covered by our terms of service.
  • ✓ The rest is our published policy. No training on your data, no recordings without written consent, and a clinician signing every note are how CoralEHR works today, and they are published here so you can hold us to them.
  • ✓ A real exit. You can export your data at any time, at no charge, and leave whenever you want.

This is the answer to the pattern that already cost real money. The FTC banned BetterHelp in 2023 from sharing customers' mental-health data for advertising and ordered $7.8 million returned to consumers, after it handed users' email addresses, IP addresses, and questionnaire answers to advertising platforms despite promising to keep them private. In 2024 the FTC ordered Cerebral to pay in excess of $7 million for exposing nearly 3.2 million people's sensitive data to ad and media platforms including TikTok, Snapchat, and LinkedIn, and for mailing roughly 6,000 postcards that revealed patients' diagnoses. Those were promises made in a privacy policy and not kept.

On compliance, the honest line: CoralEHR is HIPAA compliant and signs a BAA with every practice. We don't hold SOC 2 or HITRUST today, and we will never present a certification as held before it is earned.

Why this matters now

The appetite for the thing we're promising never to build is real and funded. Slingshot AI launched Ash, billed as "the first AI designed for therapy" and built on a special-purpose foundation model for psychology, on roughly $93 million raised. When Northern California Kaiser Permanente therapists asked for the same contract language their Southern California colleagues won in 2025, that AI's role "is not to replace but to assist" them, the union says the company declined, part of why up to 2,400 clinicians struck on March 18, 2026 (Kaiser disputes the characterization, saying AI will not replace human assessment). And a Stanford study found that when a user signaled suicidal intent, leading "therapy" chatbots responded by naming the nearest tall bridges.

Read that last one carefully. It is about chatbots posing as the therapist, which is the exact thing our pledge promises never to build. The promises above describe AI that helps with notes and admin, and we keep those two subjects apart on purpose. The full argument lives at /pledge/.

How each AI feature works

Every feature is admin work. You review and sign, and the clinical judgment stays yours. Here is what each one does.

See all AI features

Frequently asked questions

Does CoralEHR train AI on my clients' notes or sessions?

No. Your clients' data, including notes, transcripts, assessments, messages and any other PHI, is not used to train, fine-tune or evaluate any shared or foundation model, whether it is identifiable, de-identified or aggregated. We don't do it today, and we won't start without your explicit, opt-in consent, which you can withdraw. We would tell you in advance.

Does the AI keep recordings of my therapy sessions?

No. The notes assistant never stores audio, and transcripts are deleted when you sign the note, or after 7 days if you don't. Sessions are recorded only for supervision, and only with the client's written consent. Those recordings stay inside CoralEHR, can't be downloaded, and are deleted on a schedule, or right away if the client withdraws consent.

Does CoralEHR sign a Business Associate Agreement (BAA)?

Yes. The practice owner signs it inside CoralEHR, and the current text is published at /baa/. A BAA is the legal minimum for handling PHI, and the rest of this page covers what we do beyond it.

Does AI ever diagnose or decide on treatment?

No. The AI finds, drafts and summarizes, and a licensed clinician reviews and signs everything before it enters the chart. Its instructions bar it from diagnosing, rating risk and recommending treatment. Treatment-plan suggestions are coming, and when they ship each item will be a suggestion for the clinician to accept, edit or reject.

Is CoralEHR SOC 2 certified?

No. CoralEHR is HIPAA compliant and signs a Business Associate Agreement with every practice. We don't hold SOC 2 or HITRUST today, and we never claim a certification we don't have.

What happens to these promises if CoralEHR gets acquired?

The Business Associate Agreement you sign with us governs how your clients' information is protected, and it does not depend on who owns CoralEHR. The other promises on this page are our published policy today. Whatever happens, you can export your data at any time, at no charge.

How is this different from SimplePractice's AI?

On its own June 3, 2026 de-identification page, SimplePractice states that de-identified transcripts will be used to improve Note Taker and related AI features, and on its transcript-retention FAQ it states that retained transcripts and PHI are not sold. CoralEHR's promise is simpler. Your data is used only for your own practice, and we don't de-identify it to improve our tools either.

Which AI model does CoralEHR use, and is it barred from training on my data?

CoralEHR's AI runs on Anthropic's first-party Claude API through one fixed provider path, so your data is never routed to other model providers. Under our agreement, your inputs are not used to train the model. Your note content lives in the client's chart under our BAA, and it isn't sold, shared for advertising or fed to a foundation model.

See it for yourself

Book a demo to see where your data goes.