HIPAA Compliant Is Not The Same As Confidential: Simple Practice Just Proved the Point

Earlier this year, I wrote about why AI threatens confidentiality in therapy and why I won’t be using it in my clinical work. At the time, I was hearing a lot of clinicians talking about HIPAA-compliant AI tools through their EHR systems and how these were different from consumer AI like ChatGPT or Claude because they were designed for healthcare workers. The assumption was that as long as the tool was marketed for clinical use, the confidentiality risk wasn’t a problem.

Today, Simple Practice, one of the most widely used EHR platforms in private practice, sent out a notice that proved otherwise.

What Simple Practice Announced

The email, signed by Simple Practice’s Head of Clinical Strategy, announced that starting June 16, 2026, the platform will begin retaining session transcripts generated by their AI Note Taker feature. The language is carefully wrapped in reassurances about HIPAA compliance and de-identification. But buried in that reassuring language are two disclosures that should alarm clinicians.

First: they retain session transcripts. Most clinicians assume that once an AI tool generates a note, the underlying transcript—the actual record of what was said in session—is erased. That’s not what’s happening. In order for AI to “listen in” and take notes, it has to record the session. And that recording is being retained as a transcript.

Second: those transcripts are being used to train their AI. Simple Practice states openly that the reason for retaining de-identified, de-coupled transcripts is “to continuously improve existing and upcoming AI features.” In their own words, the retained session material helps with:

“Understanding clinical nuance and context. Improving note accuracy and clinical language. Reducing biased or inaccurate interpretations. Recognizing therapeutic progress and client sentiment.”

In other words, actual session material—what patients said in the room they believed was private—is being fed into a system designed to get better at processing future sessions. Simple Practice assures users that this data is de-identified and de-coupled, meaning names and identifiers are scrubbed and the transcript can’t be linked back to a specific clinician or client. But de-identified is not the same as gone. The content of what was said still exists and is still being used.

HIPAA Compliance Is Not the Same as Confidentiality

This is where it’s critical to understand the difference between HIPAA and confidentiality, because they are not the same thing.

Confidentiality is a foundational principle of the therapeutic relationship. It means that anything said in session remains private. The clinician cannot disclose it to others without going through proper channels. There are well-established limits to confidentiality such as mandated reporting of child abuse or the duty to warn when someone is in danger, but these exceptions generally exist to protect people.

HIPAA is the legal framework that regulates how practices handle protected health information. It includes exceptions for things like treatment coordination, billing, public health reporting, mandated abuse reporting, duty to warn, and court subpoenas — some protective, some operational.

But what Simple Practice is doing (and what I suspect other healthcare AI’s are or will be doing) is insidious because once session material is de-identified under the Safe Harbor method — with all 18 specified identifiers removed — it is no longer considered protected health information. HIPAA stops applying to it entirely. There are no restrictions on how that data can be used, stored, or shared. The content of therapy sessions has been moved outside the law's protection altogether.

Don’t take my word for it. HHS, the federal agency that administers HIPAA, is explicit about this here:

“Both methods, even when properly applied, yield de-identified data that retains some risk of identification. Although the risk is very small, it is not zero, and there is a possibility that de-identified data could be linked back to the identity of the patient to which it corresponds. Regardless of the method by which de-identification is achieved, the Privacy Rule does not restrict the use or disclosure of de-identified health information, as it is no longer considered protected health information.” (emphasis mine)

Even HHS acknowledges the risk of re-identification isn't zero. Simple Practice simply cannot promise that identification of the patient isn't possible from their own words just because it has scrubbed the material of the specific information that HIPAA would consider identifiable information. And once de-identified it is at even more risk of misuse because it no longer has the legal protections associated with HIPAA.

Why This Should Concern You Whether You’re a Clinician or a Patient

What makes this different from other HIPAA exceptions is how voluntary and completely unnecessary this is. No one is being protected by it. A clinician chooses to use AI for note-taking for convenience—sometimes without fully informing the patient of what that entails—and in doing so, creates a pathway for actual session material to be retained and used by a technology company.

Simple Practice’s email frames transcript retention as something that will be “enabled by default” for anyone using Note Taker, with opt-out options available. A clinician who doesn’t read the email carefully, or who assumes HIPAA compliance means their patients’ confidentiality is protected, may never realize they need to opt out. And their patients may never know this is happening at all.

Legal Compliance as a Loophole

This is not unique to Simple Practice. It’s a pattern across the tech industry: adhere to the legal framework while subtly doing something that undermines user privacy and mines their data for profit. De-identification becomes the mechanism by which confidentiality is technically maintained while the substance of what was confidential—the actual words spoken in a therapy session—gets repurposed and loses its protection.

Because it’s done legally, it’s unlikely to raise the alarm that it should...unless people understand that in the tech world, a company can adhere to the letter of the law while betraying the spirit of it.

I wish we could trust all providers to understand these risks and protect their patients accordingly. I’m sure that many conscientious and attentive providers will do exactly that. But there are enough clinicians who are either unfamiliar with the technology they’re using or who simply don’t look closely enough at what they’ve agreed to.

What You Can Do

If you’re a clinician: Don't use AI with clinical material. Period. The convenience isn't worth the confidentiality leak, and as Simple Practice just demonstrated, even platforms marketed as HIPAA-compliant can change their data practices at any time. You can't un-share what's already been processed. The safest way to protect your patients' confidentiality is to keep their session material out of these systems entirely.

If you’re a patient: You have every right to ask your providers whether they use AI in any aspect of your care—note-taking, session recording, treatment planning, anything. If they do, ask to opt out and do so in writing. You shouldn’t have to be the one protecting your information, but right now, you have to be.

The Room Should Stay Private

Therapy depends on the belief that what happens in the room stays in the room with very few exceptions. That belief is what allows people to say the things they’ve never said out loud, to explore the parts of themselves they’re most afraid of, to be honest in ways they can’t be anywhere else.

When session material gets retained and used for product development, something essential about the therapeutic container is broken. HIPAA compliance isn’t the same as protecting confidentiality in this day and age. And right now, it’s on all of us—clinicians and patients alike—to know the difference.