Image: Episode 501 artwork for Modern Therapist's Survival Guide. A close-up image of a wood file cabinet appears in the background. Large white text reads
Google PodcastsApple PodcastsStitcherSpotifySubscribe on AndroidShare Listen in a New WindowDownloadSoundCloudLeave a ReviewSubscribe via RSSKeep me updated!Player Embed

Is Your Client Data Safe with AI? What Therapists Need to Know About HIPAA, BAAs, and Session Transcripts – An Interview with Ali Hartley, JD

You probably know you need a business associate agreement with your EHR, your AI note taker, and any other software that handles client information. But do you know what that agreement actually protects, where session content travels once an AI scribe starts listening, or what happens to a transcript after the note is signed? Most therapists are feeling uneasy about data privacy in the age of AI, but some are also terrified that this data is creating their replacements.

Curt and Katie take both concerns to Ali Hartley, JD, Chief Legal Officer at SimplePractice. Ali explains how HIPAA applies to AI documentation tools, why a BAA is necessary but never sufficient, and what clinicians should ask before adopting any tool that handles client data. She also clears up the difference between training an AI model and refining the prompts that guide a product, including what SimplePractice actually does with de-identified session transcripts. This episode is part of our partnership with SimplePractice, and it doubled as our own due diligence as SimplePractice customers.

This is a continuing education podcourse.

Transcript

Click here to scroll to the podcast transcript.

(Show notes provided in collaboration with Otter.ai and Claude AI.)

About Our Guest: Ali Hartley, JD

Image: headshot of Ali Hartley

 

Ali Hartley, JD, is the Chief Legal Officer at SimplePractice and an accomplished general counsel and healthcare attorney with nearly two decades of experience in corporate law, regulatory compliance, and people leadership. She helps SimplePractice remove administrative burdens for clinicians while enabling responsible innovation and protecting sensitive healthcare data.

 

In This Podcast Episode: HIPAA Compliance and Data Privacy for Therapists Using AI Tools

Ali explains how HIPAA’s Privacy, Security, and Breach Notification Rules apply to AI tools, where protected health information is most vulnerable in an AI scribe workflow, and what a BAA does and doesn’t guarantee. She compares HIPAA’s Safe Harbor and Expert Determination methods of de-identification and separates training a large language model from refining the prompts that guide a product. Using SimplePractice as a case example, she describes how transcripts are retained and decoupled from client and clinician records, and how clinicians can opt out globally, by client, or by session. Curt, Katie, and Ali also compare native EHR integrations with browser extensions and copy-and-paste workflows, and look ahead to the risks therapists should watch as AI tools multiply.

Learning objectives and full course details are available in the Modern Therapist Learning Community.

Key Takeaways for Therapists: BAAs, De-identification, and Safer AI Documentation

“A BAA does not encrypt your data, does not prevent a breach. It doesn’t mean that the vendor has a good security infrastructure… A BAA is a piece of paper. It’s a legal backstop. It’s not a technical one.”

— Ali Hartley, JD

  • HIPAA is technology neutral. If an AI tool touches PHI, HIPAA governs that interaction. Meeting the Privacy, Security, and Breach Notification Rules takes more than a strong password: a documented risk assessment, written policies, encryption at rest and in transit, audit logs, and a plan for when something goes wrong, revisited at least once a year.
  • A BAA is the compliance floor. You need one with every vendor that touches PHI, including the small plug-ins and extensions that seem harmless. A signed BAA starts the conversation. Independent security certifications, breach history, and subprocessor agreements tell you whether the vendor can back it up.
  • Every step in an AI workflow is a door. Audio capture, transcription, note generation, and what happens to the raw audio and transcript afterward are each a point where data can leak. Ali’s question for any tool is whether every one of those doors stays locked.
  • De-identification has two legal paths. Safe Harbor removes 18 specified identifiers. Expert Determination relies on a qualified expert certifying that re-identification risk is very small. Decoupling goes further than HIPAA requires by severing the link between de-identified text and any clinician or client record.
  • Refining prompts is different from training a model. Training a base LLM changes the model’s foundational weights, and the data can’t be pulled back out. Using de-identified transcripts to check note quality and adjust the instructions that guide a tool leaves the underlying model unchanged.
  • Where the AI lives changes your risk. Native EHR integration keeps data inside one security perimeter. Browser extensions can read what is on your screen and update without notice. Copy-and-paste exposes your clipboard and leaves no audit trail.
  • You own what you sign. AI-generated notes need the same clinical review as notes you write yourself, and Ali encourages ongoing, transparent conversations with clients about how you use AI in your practice.

“When an AI generates a clinical note and you sign it, you own it. If that note contains an error or an inaccurate clinical impression or a mischaracterized statement from your client, you are professionally and legally responsible for that.”

— Ali Hartley, JD

Why This Matters: Vetting the Tools That Touch Client Data

As the covered entity, you are responsible for the entire data chain, from the vendor you sign with to every subprocessor behind it. Ali’s advice is to map your data flow from end to end and ask every vendor the same questions:

  • Will you sign a BAA, and can I read it before I sign up?
  • Do you publish how you process data, and do your subprocessors sign BAAs too?
  • Do you hold independent security certifications, and have you had any publicly reported breaches?
  • Is data encrypted in transit and at rest, and is it stored onshore?
  • Which large language model is the tool built on, and does data stay walled off or fall under a zero data retention policy?
  • Was the tool built for behavioral health, and can I review and edit everything before it’s finalized?

If you use tools outside your EHR, apply the minimum necessary standard and strip names, dates of birth, and other direct identifiers before anything leaves your system.

If all of this feels like a lot, Ali’s reassurance is that you don’t have to become a privacy expert or a lawyer to protect your clients. Ask the hard questions, read the BAA, and be intentional about the tools you use and the partners you trust.

Resources on HIPAA, Business Associate Agreements, and AI Data Privacy

We’ve pulled together resources mentioned in this episode and put together some handy-dandy links. Please note that some of the links below may be affiliate links, so if you purchase after clicking below, we may get a little bit of cash in our pockets. We thank you in advance!

Continuing Education Information

Hey modern therapists, we’re so excited to offer the opportunity for 1 unit of continuing education for this podcast episode – Therapy Reimagined is bringing you the Modern Therapist Learning Community!

Once you’ve listened to this episode, to get CE credit you just need to:

Once completed, your CE certificate will appear in your profile and can be downloaded for your records.

You can find this full course here: https://learn.moderntherapistcommunity.com/courses/is-your-client-data-safe-with-ai-what-therapists-need-to-know-about-hipaa-baas-and-session-transcripts

Continuing Education Approvals

When we are airing this podcast episode, we have the following CE approval:

Therapy Reimagined is approved by the California Association of Marriage and Family Therapists to sponsor continuing education for LMFTs, LPCCs, LCSWs, and LEPs (CAMFT CEPA provider #132270). Therapy Reimagined maintains responsibility for this program and its content. Courses meet the qualifications for continuing education credit as required by the California Board of Behavioral Sciences. Please check with your licensing board to confirm eligibility.

Please check back as we add other approval bodies: Continuing Education Information including grievance and refund policies.

References Mentioned in This Continuing Education Podcast

Relevant Episodes of MTSG Podcast

Meet the Hosts: Curt Widhalm & Katie Vernoy

Picture of Curt Widhalm, LMFT, co-host of the Modern Therapist's Survival Guide podcast; a nice young man with a glorious beard.Curt Widhalm, LMFT

Curt Widhalm is in private practice in the Los Angeles area. He is the cofounder of the Therapy Reimagined conference, an Adjunct Professor at Pepperdine University and CSUN, a former Subject Matter Expert for the California Board of Behavioral Sciences, former CFO of the California Association of Marriage and Family Therapists, and a loving husband and father. He is 1/2 great person, 1/2 provocateur, and 1/2 geek, in that order. He dabbles in the dark art of making “dad jokes” and usually has a half-empty cup of coffee somewhere nearby. Learn more at: http://www.curtwidhalm.com

Picture of Katie Vernoy, LMFT, co-host of the Modern Therapist's Survival Guide podcastKatie Vernoy, LMFT

Katie Vernoy is a Licensed Marriage and Family Therapist, coach, and consultant supporting leaders, visionaries, executives, and helping professionals to create sustainable careers. Katie, with Curt, has developed workshops and a conference, Therapy Reimagined, to support therapists navigating through the modern challenges of this profession. Katie is also a former President of the California Association of Marriage and Family Therapists. In her spare time, Katie is secretly siphoning off Curt’s youthful energy, so that she can take over the world. Learn more at: http://www.katievernoy.com

A Quick Note:

Our opinions are our own. We are only speaking for ourselves – except when we speak for each other, or over each other. We’re working on it.

Our guests are also only speaking for themselves and have their own opinions. We aren’t trying to take their voice, and no one speaks for us either. Mostly because they don’t want to, but hey.

Join the Modern Therapist Community:

Linktree

Patreon | Buy Me A Coffee

Podcast Homepage | Therapy Reimagined Homepage

Facebook | Facebook Group | Instagram | YouTube | LinkedIn | Substack

Consultation services with Curt Widhalm or Katie Vernoy:

The Fifty-Minute Hour

Connect with the Modern Therapist Community:

Our Facebook Group – The Modern Therapists Group

Modern Therapist’s Survival Guide Creative Credits:

Voice Over by DW McCann https://www.facebook.com/McCannDW

Music by Crystal Grooms Mangano https://groomsymusic.com

Transcript for this episode of the Modern Therapist’s Survival Guide podcast (Autogenerated):

Transcripts do not include advertisements just a reference to the advertising break (as such timing does not account for advertisements)

… 0:00
(Opening Avertisement)

Announcer 0:00
You’re listening to the Modern Therapist Survival Guide, where therapists live, breathe, and practice as human beings. To support you as a whole person and a therapist, here are your hosts, Curt Widhalm and Katie Vernoy.

Curt Widhalm 0:15
Hey, modern therapists! We’re so excited to offer the opportunity for one unit of continuing education for this podcast episode. Once you’ve listened to this episode, to get CE credit, you just need to go to moderntherapistcommunity.com, register for your free profile, purchase this course, pass the post test, and complete the evaluation. Once that’s all completed, you’ll get a CE certificate in your profile, or you can download it for your records. For a current list of our CE approvals, check out moderntherapistcommunity.com.

Katie Vernoy 0:47
Once again, hop over to moderntherapistcommunity.com for one CE once you’ve listened. Woohoo!

Curt Widhalm 0:54
Welcome back, modern therapists. This is Modern Therapist Survival Guide. I’m Curt Widhalm with Katie Vernoy, and we’re going to be joined by Ali Hartley, Chief Legal Officer of SimplePractice, in a little bit. But this is the podcast for therapists about the things that go on in our practices, and this is another one of our continuing education eligible episodes where we’re going to be talking about how our laws interact with the technologies that we use and things that we should be aware about as clinicians. And Katie, this is a potential for a episode that could get very dry, very boring. It’s not. There’s a lot of great information in this episode, but this is an important discussion to have.

Katie Vernoy 1:48
We reached out to SimplePractice as part of our partnership to determine how do we make sure, as part of our ongoing vetting process, both as SimplePractice customers, but because we’re recommending them to you, how do we make sure that their data privacy policies make sense to us? That we’re remaining HIPAA compliant. What is the mysteries around de-identified, decoupled transcripts. Are they training a model? What does that mean? And we’ve had a couple of conversations with Ali about this. And so the the one that we’re going to share with you is really it’s continuing education for you, and it’s also due diligence for us. Which I feel completely satisfied, and I think it’s very interesting for anyone that’s concerned. I would take a good listen to this one because it’s it’s got a lot of information, very transparent about how SimplePractice handles our data, and what you should look for from other other vendors and other tools you might be using, whether they’re AI tools or or otherwise. And so, definitely take a listen if you’re you’re wanting to make sure that you’re you’re keeping your clients’ privacy safe and you’re complying with HIPAA.

Curt Widhalm 3:12
Since this is a continuing education episode, we do have some learning objectives. Our first one is differentiate between three core HIPAA data handling mechanisms, including business associates agreements, safe harbor de-identification standards, and data decoupling protocols. Number two is evaluate three security vulnerabilities, memory leak risks, and data pathways across workflows. And our third one is formulate a clinician-centered vetting framework to audit third-party software tools for your practice.

Katie Vernoy 3:50
And just and in case it’s not clear, this is a CE episode that will qualify for law and ethics. So you’re going to accomplish a lot just listening to this next conversation.

Curt Widhalm 4:05
Remember when the biggest technological crisis in private practice was accidentally hitting reply all to an email? Simpler times. Fast forward to today, and it seems like half the profession is terrified that the new AI scribes are secretly building a digital clone to replace them by Friday, and there’s other groups that are just casually pasting unencrypted progress notes into sketchy browser extensions they downloaded at 2 a.m. So between BAA’s, safe harbor standards and the existential dread of machine learning, trying to keep your practice HIPAA compliant feels less like running a business and more like defusing a bomb in a web browser. So today, Katie and I are joined by SimplePractice’s Ali Hartley to separate the real data privacy risks from the sci-fi paranoia. We’re answering the big questions such as what does a BAA actually protect? What really happens to your session transcripts, and are we accidentally teaching robots how to do active listening? So, thanks for listening to us. This is a CE eligible episode, and , thank you so much for joining us and sharing all of your expertise as well.

Ali Hartley 5:18
Hey, thank you so much for having me here today, I’m honored to be on your podcast.

Katie Vernoy 5:23
We’re excited to get into this conversation, and as a disclosure, this is part of our partnership with SimplePractice. I’m going to go ahead and ask you the question we ask all of our guests when we get started, which is, “Who are you, and what are you putting out into the world?

Ali Hartley 5:36
I love that question. Thank you. So, hi everyone. I’m Allie Hurley. I’m the Chief Legal Officer at SimplePractice. SimplePractice builds practice management platform and EHR for over 260,000 therapists, counselors, and wellness professionals that they can run their private practices. My job, I guess, I would say in a sentence, is to make sure that those practitioners can adopt powerful technology without having to worry about whether they’re putting their clients at risk. I love telling people that I’m a lawyer who chose to work at a company where the product generally changes lives. You know, when a therapist spends two fewer hours a night on documentation, that’s two hours they get up with their family, or two hours they can see another client who’s been on a wait list. You know, we all know obviously that the mental health crisis, you know, access crisis across this country is real. So an administrate an administrative burden is one of the reasons practitioners burn out and leave the field. So when I think about what I bring to this conversation, it’s an intersection of privacy law, health tech, private product development, a deep belief that compliance should be an enabler and not a barrier. So when people hear chief legal officer at a health tech company, they often assume I spend all day reading regulations in a windowless office, and some of that is actually accurate. But but most of my time is spent figuring out how to help innovation in a way that keeps our practitioners safe. So my team’s job really is to be what I would say guardrails on a highway.

Curt Widhalm 7:08
So when we were setting up this episode, we Ali and I bonded very much over our love for HIPAA, and you know, there’s nine of us in the world who really do that. But just to make sure that we are all on the same page about things, I know when I taught law and ethics classes for therapists that it seems like a lot of the discussions for HIPAA is around the 18 identifiers that we get, and so that’s where we see a lot of therapists talk about, oh, you’re violating HIPAA by sharing somebody’s email address or sharing identification data about them. But I want to maybe talk a little bit of a brief overview. That number one, that’s not the only part of HIPAA, and what it really means as far as it pertains to our conversation here today.

Ali Hartley 7:58
HIPAA really, you know, is is can be overwhelming, honestly. It’s a giant set of of regulations, and and Curt, as you said, it goes well beyond sort of thinking about identifiers. There are actually sort of three big interlocking rules within HIPAA that are worth noting. The first is the privacy rule, and that governs who can see PHI, under what circumstances, and with what authorization. The second is the security rule, which requires you to have administrative, physical, technical safeguards in place to protect electronic protected health information. And then the third is the breach notification rule, which means that if something goes wrong, you have an obligation to notify affected individuals. Sometimes the Department of Health and Human Services, and in some cases the media. So for a solo or small group therapist, you know meeting HIPAA standards goes well beyond things like thinking about identifiers or having a strong password. You should have a documented risk assessment, which is an honest evaluation of where PHI lives in your practice and what could go wrong. You should have written policies within your practice on who can access what and for what purposes, and then you should ensure that encryption, you have encryption for the data at rest and in transit. You should have audit logs so you can see who access what records and when, and then you need a plan for what happens when laptops get stolen, or a vendor gets breached, or something like that. You need to revisit all this regularly. I would recommend on an annual basis because your risk profile changes every time you adopt a new tool. All this to say is, you know, you are a therapist running your own practice out of a small office, and you’re the CEO, and you’re the billing department, and you’re the IT department, and of course you’re the clinician. HIPAA doesn’t necessarily care that you’re one person. You still need all those things that I talked about, and this sounds overwhelming. Honestly, it’s one of the main reasons we built SimplePractice the way we did is, you know, we wanted to bake in that compliance infrastructure into our platforms so that therapists don’t have to become part-time lawyers or security analysts or experts in HIPAA. So when you use SimplePractice, you know there are there are built-in access controls, audit logging, encryption to protect your data both at rest and in transit, and without requiring that sort of manual configuration that you would have to think about if you weren’t using SimplePractice. So we’ve really done the heavy lifting on those technical safeguards, so that practitioners can focus on the clinical work.

Katie Vernoy 10:34
Just a reminder: this is a continuing education podcast. Feel free to roll that back and remind yourself of those three interlocking pieces of HIPAA. I was always told, like, if I just have a BAA, everything is fine. And then now there’s obviously these additional pieces that you talked about, but it does seem like even the idea of BAA seems suspect, or how do I know, and what’s happening, and you know all these things. And so, what does HIPAA mean for what it is now? And and maybe for folks who don’t know, what’s a BAA, and how does that all fit in?

Ali Hartley 11:14
So let’s let’s talk about like how HIPAA fits into this new world of AI, right? HIPAA wasn’t written with AI or really any particular technology in mind, but it absolutely applies to it, right? So the law is technology neutral by design. It doesn’t say protect PHI stored on a fax machine differently than protect PHI processed by AI or protect PHI stored in your computer, like it, it is all like the law governs all of that, regardless of the type of technology. So if AI touches PHI, HIPAA governs that interaction. Period. What changes with AI is is kind of the attack surface and the data flow complexity. So let’s walk through like what actually happens when an AI scribe processes a therapy session because the vulnerability points become obvious or a little bit more obvious once you see that change. So the first step is the capture, right? Audio is recorded either locally on a device or streamed to a server, and the vulnerability point here you start to think about: Is the audio stream encrypted in transit? Is the microphone access limited to the authorized application, or could other processes on the device intercept it? Step two is the transcription. So when the audio is converted to text, often by speech-to-text model, there is a vulnerability point there. You know, did the data just leave the vendor’s infrastructure and go to a subprocessor, or is there a BAA in place with the subprocessor? And we’ll talk about BAAs obviously in a second. The third step is sort of that clinical note generation. So the transcript is then processed by AI that interprets the clinical context, generates a structured note. So the vulnerability point there is, you know, is the model hosted on the same infrastructure? Is there EHR? Is it calling to an external API? Is there a transcript or a note cached somewhere along the way? And then the fourth is sort of the the generated note is then saved into the client’s record. But what happens to the raw transcript and the audio? And we’ll talk about that, I believe, a little bit later. You know, are they deleted? Are they retained? What are they used for? So at SimplePractice, you know, our AI scribe operates within our existing HIPAA compliant infrastructure. So, you know, session audio is deleted immediately. The transcript is stored with limited access only until either the note is signed and locked, or seven days, whichever has come first. And after that, transcripts are either deleted or fully de-identified and decoupled, depending on the clinician’s personalized choices. Then data never, importantly, leaves our secure environment. So what that means is, each of those vulnerability steps I just described is handled within a single security perimeter. So at every you know each of those sort of door like steps, the vulnerability steps, I would say is a door. So the question is whether is the door always locked? Is every single door locked along the way? Right. So the thing that like I worry about a lot when I think about these types of scenarios for therapists is, and I think we’ll talk about this a little bit later too. Is sort of like downloading free Chrome extensions that promise to transcribe, or using multiple tools across multiple browsers and things like that. These therapy sessions obviously contain the most intimate details of someone’s mental health struggle, and you know you you got to have a strong obligation to protect that, and that obligation comes with the BAA. So this sort of goes into your second second question, which is like, what is a BAA? Like what does it actually cover? And so a BAA is a Business Associate Agreement, and what that is is it’s a legal contract between a covered entity, which is all the therapists, and a third party that may create, receive, or maintain or transmit PHI on behalf of that covered entity. So under HIPAA, the law requires that you need a BAA with every single vendor that is in your chain, every single vendor that you share PHI with, including if you were to put a Chrome extension or plug in into your into you know into your practice. Sort of a like a everyday way I would describe a BAA to non lawyers is pinky promise with teeth. So it’s a vendor saying, “I will protect this data. I will use it only for the purposes that you’ve authorized, I will tell you if something goes wrong, and if I screw up, there are real consequences to me screwing up,” right? So it is that sort of it’s that legal contract, but there are real world and very heavy consequences for screwing up. The critical part, and what I want really people, you know, people to take away in this conversation is is a BAA is necessary. It’s required by the law, but I would say it’s not sufficient. So what a BAA does is it does create that contractual obligation and it defines the permitted uses of the PHI and establishes breach notification duties, all that stuff, and it gives you the legal remedy if a vendor fails. But what a BAA does not do is it does not encrypt your data, does not prevent a breach. It doesn’t mean that the vendor has a good security infrastructure. It doesn’t mean they’re actually following through on what they promise. So a BAA is a piece of paper. It’s a legal backstop. It’s not a technical one.

… 16:41
(Advertisement Break)

Ali Hartley 16:42
When you use SimplePractice’s AI features, you’re already covered under SimplePractice’s existing BAA. No separate contracts are required. You don’t need to negotiate additional vendor contracts, anything like that. You know, our team already handles the BAA negotiations with our partners or our vendors, including provisions like model retraining restrictions, like prohibiting anyone from using PHI to train models or anything like that. When I say it doesn’t stop with just the BAA, we take that really seriously. We don’t stop at the BAA. You know, we back everything we do with high trust certification, annual audits, advanced encryption, regular penetration testing done by independent security firms, and honestly, around the clock security operations center. So, when I talk to a therapist and I tell them, I tell them, if a BAA, if a vendor won’t sign a BAA, run. But if they will sign one, that’s just the beginning of the conversation, right? You need to follow up with understanding whether they have, you know, the vendor has independent security certifications. Ask to see that documentation. You know, ask about sub processors and whether they have BAAs as well. So it is, I would say, it’s an ongoing conversation that therapists need to be having with their vendors.

Curt Widhalm 17:58
Pulling a couple of practical things out of what you’re talking about. You’re talking about Chrome extensions. So if somebody has something such as Grammarly, you know, a popular extension that makes us all sound smarter when we write, that we don’t have BAA’s with them, and that has a vulnerability for clinicians if that’s overlaying on top of where they’re writing their clinical notes.

Ali Hartley 18:20
I think there’s sort of like different architectures we can talk about, right? So the first would be like native EHR integration, which which is what SimplePractice does, and you know it’s an AI tool built directly into your practice management platform. So data never leaves that security perimeter that we’ve talked about and already vetted that you’ve already vetted.

Curt Widhalm 18:40
So when I’m in Simple Practice, everything is operating within Simple Practice.

Ali Hartley 18:45
Exactly. Yeah. So that’s what I would call like a native EHR integration. And then there are what you just mentioned, which is like Chrome overlays and browser extensions. So the way this works is that the tools sit on top of your browser and interact with your EHR through that browser environment. So the security concerns are are pretty specific and pretty real here. So number one is like browser extensions can access the structure of the web page you’re viewing. So if you have your EHR open on that same browser, the extension may access may have access to data displayed on the screen, even data you’re not intending to share with that tool. The other piece is that browser extensions can include analytics or advertising trackers that send data to parties you’ve never heard of, which is called third-party tracking. So, if the browsing data includes your EHR session, that’s a potential PHI exposure you didn’t consent to. I would say the third sort of big risk is like browser extensions update automatically, and so a tool that like you may look at once and is safe today could push an update tomorrow that changes the way it handles data, and you’d never know unless you’re like reading change laws. Which which most people aren’t, I would say.

Curt Widhalm 20:03
You had also talked about the questions that people should be asking with BAA’s and talking about sub processors or third party AI foundations. For most clinicians, I would imagine that they would walk out of what we’ve talked about so far, and go to some place such as SimplePractice and say, “I’m supposed to ask about subprocessors and third-party AI foundations. What kinds of questions should they be asking to hopefully be able to understand what what is going on underneath the hood?

Ali Hartley 20:37
Yeah, so I would I would say there’s there’s a couple fairly easy places to start to see, like you know, is a vendor doing what they say they’re doing, right? Which is they’re saying we have a BAA. Is that BAA accessible? Can you read it online? Can you find it easily? And then I would say, secondly, like, does that vendor publish information about how they, you know, process data and and what they do and do they share do they share public data commitment that they have subprocessors and those subprocessors are also required to sign BAA’s. Another thing you can do is like search for past breaches. You know, has the vendor had security breaches? Right. That will give you a good sense of what their security architecture is, and how how sort of strong it is. For SimplePractice, you know, we have a trust center online where you can go and get a lot of information about our specific security practices. You know, our BAA information about subprocesses, things like that. So we have, and we have our privacy policy published. Everything is sort of publicly available, and the reason we do that is we want to be really transparent for our customers, so that they understand, you know, how we take our obligation extremely seriously to protect PHI.

Katie Vernoy 21:55
SimplePractice has the the trust center, but for information that’s not held by the company, like breaches or or other types of information. Are there typical places where you could find out if there’s been breaches in a at a company where you’re wanting to store some data?

Ali Hartley 22:15
Breach information, you know, HIPAA does require larger breaches to be shared with the Department of Health and Human Services, and then if it’s a large enough breach, it has to go to the media. So often, just a Google search can do it, right? Like if you just Google has you know so and so had a you know PHI HIPAA security breach? That information might pop up.

Katie Vernoy 22:39
Okay.

Ali Hartley 22:39
Depending on the size of the breach.

Katie Vernoy 22:40
So it should be it should be pretty findable.

Ali Hartley 22:43
Yeah.

Katie Vernoy 22:44
Okay.

Ali Hartley 22:44
Our legal team, you know, when we work with vendors, we do a full security sweep of every vendor we work with as well. We do that same thing. You know, we check to see if they’ve had any breaches. You know, what their security practices are. We require a lot of documentation from all of our vendors around their their specific security practices, and so you know that stuff is definitely findable.

Katie Vernoy 23:07
So, if someone is wanting to work with a particular application that that that should be HIPAA compliant, what are the red flags? You said if they won’t sign a BAA, run. Are there other red flags that that you want folks to to pay attention to?

Ali Hartley 23:24
Yeah, I think if they, you know, if they’re not, you know, if they’re not publicly posting their BAA, you know, some some people don’t want that public because there there might be some sort of shady information in it, shady shady stuff in their BAA around what they’re doing with PHI and and how, you know, if they’re not publicly, I wouldn’t say it’s a red flag like run away, but like it’s it’s a thing to look for. They publicly putting out there what they do with data, and they being transparent about it. You know, I think folks who are not being transparent says a lot. You know, it’s a it’s not, I again, it’s not a run thing because some people might they might like you know, if you reach out to a vendor, if it’s not public, but if you reach out to the vendor, they might be very willing to share with you that information. So I’m not saying like just because it’s public doesn’t mean doesn’t mean you should run, but ask the questions. You know, if they’re willing to have a dialog with you, that’s a really good sign.

Curt Widhalm 24:16
A lot of the discussion that we’re hearing about SimplePractice and about AI scribes and all of the things that go on under the hood is really around the de-identification that goes on with this. And de-identification is allowed within HIPAA, but can we talk a little bit about what this actually means? What it actually looks like?

Ali Hartley 24:41
Totally. So, under HIPAA, health information is considered de-identified, can no longer identify an individual, and cannot reasonably use be used to identify one. Under HIPAA, there are two different methods to get the de-identification. First is called the safe harbor, and I would say this is the more prescriptive approach. So, the safe harbor requires that 18 specific identifiers get removed. So, it is names, dates, geographic data, phone numbers, fax numbers, email addresses, social security numbers, medical record numbers, health plan beneficiary numbers, account numbers, certificate license numbers, vehicle identifier. I mean, like it was on and on, including URLs, IP addresses, things like that, biometrics. So what the law says is, if you strip all those 18 categories and have no actual knowledge that the remaining information could identify someone, the data is considered de-identified under HIPAA. The second method is what’s called the expert determination method, and this is more what I would say is a more flexible method. So what that means is that there is a qualified statistical expert analyzes the data and then certifies that the risk of re-identification is very quote unquote very small, given the anticipated recipients and the data’s context. So this method actually lets someone retain more granular data, like partial dates or broader geographic information, as long as the expert can determine that that minimal re-identification risk is is is like minimal. So SimplePractice actually chose the safe harbor de-identification method plus decoupling because it is a more conservative architecture, truly, the specific use case of transcript de-identification. We actually did evaluate whether expert determination method made sense for us, but that method is typically used when someone can’t remove all 18, and we do, and that person still wants that, you know, that re-identification risk to remain small. The method also, the sort of expert determination method, might also be used when someone plans to share that de-identified data externally. And we’re actually doing the opposite in both scenarios, which is like we’re removing the HIPAA safe harbor identifiers and then severing links, you know, that decoupling piece back to the clinician that could be there. We’re totally severing that, and then we are actually keeping the, we’re not sharing de-identified transcripts outside of our technical ecosystem. So, you know, we test that approach honestly to make sure it meets those rigorous standards, and and we’ll continue to audit and test it, making sure that it continues to be that way going forward.

Katie Vernoy 27:24
Even with all of this, it sounds really good, and folks are concerned that there are things beyond those 18 specified identifiers that might make someone identifiable, whether it’s a turn of phrase or a particular topic they talk about frequently, and that there’s somehow a way to reconstitute and reconnect some of these things. Should we be worried about that? And if so, or if if not, why not? And if so, why? What what else can we do?

Ali Hartley 27:54
One of the things to think about is sort of the you know de-identified transcripts go into a large data set, a large pool that all sort of gets mixed together, right? The purpose that the key piece here, I think, is the purpose that the de-identified transcripts are being used are to improve quality of the AI features that we do. So we’re not using that data to essentially create an underlying an underlying AI model where that data could potentially resurface, right? So it’s there’s there’s no real way for a sort of like contextual story in its whole part to resurface again once something is de-identified, if that makes sense. So it doesn’t sort of come back up anywhere. It’s the data is used just to analyze quality assurance and quality improvement, as opposed to like creating like this is the type of story therapists hear, and let’s surface it again somewhere. If that makes sense.

Katie Vernoy 28:59
It does. I think the the challenge is folks worried about data breaches or those types of things. And as you’ve said, you’re you know, and SimplePractice is very clear about it being high trust certified that you have these things. So the data is secure, and what the way you’re using it is really limited to quality improvement. So I do want to talk a little bit about the quality improvement because I think folks are, you know, I think there are folks who are worried: is it being used to to train an AI therapy model? We’ve talked about this before. We can say it again. You’re not doing that. It doesn’t make sense to do that. You would lose business if suddenly there was an AI therapist that came out for for SimplePractice.

Ali Hartley 29:43
Yeah, exactly.

Katie Vernoy 29:44
But but how how are those transcripts being used? What what does that actually look like?

Ali Hartley 29:49
Yeah. So this is a really confusing area. I will be totally honest because like AI is new and it’s moving so fast and it’s it’s very confusing. So I will I will do my best to help sort of explain what training versus quality improvement means, and I’ll try to use some analogies along the way that that I hope will help clarify things. But you know I think there’s sort of two fundamentally different I would say things when people mean when they say a company is using data to train AI, and they actually have really different implications. So, the first is what I’ll say is like training a base LLM, and what this means is it means taking massive amounts of data and using it to adjust the foundational neural network weights of a large language model. So this is what like Open AI did to build Chat GPT. What Anthropic did to build Claude? Google did to build did to build Gemini, and it requires an enormous amount of data sets, enormous computing power. It changes the model at the structural level. The data becomes, in a real sense, sort of like baked into what I’d call the model’s brain, and I’m putting quotes out here. Obviously, AI doesn’t have a brain, but like that’s a is easier way to think about it. So it’s the scenario that makes therapists most uncomfortable, and understandably so, because once data is embedded to model weights, you cannot extract it. The second area is evaluating and refining prompts and product accuracy, and this is like totally different activity. So that this is where a company, you know, takes like a de-identified transcript, for example, and uses that to test whether the AI product is generating good clinical notes. You know, are the notes accurate? Do they capture the right therapeutic modalities? Do they handle different note formats correctly? And this work typically involves humans reviewing some AI outputs against source materials, score quality, or using the transcripts to tune the instructions that guide that AI’s behavior. So the transcripts are what I would say is reference material for quality assurance, meaning we use an output of quality assurance to modify feature prompts, not as raw material for building an architecture. So let me let me use an analogy that I hope will help a little bit here. Think about it like a car manufacturer. So training a base model is like building an engine from raw steel, refining prompts is like adjusting the mirrors, the seat position, or the steering wheel, whatever, so that car drives well for a specific person. So both involve involve working on a car, right, or training AI, or whatever you want to call it. But they’re actually really fundamentally different activities with fundamentally different implications for what happens to those raw materials. So most companies in the behavioral AI space, including SimplePractice, are doing that second thing I talked about, which is using de-identified data to evaluate and improve product quality. SimplePractice, we are not using transcripts to update, retain, or modify, you know, an underlying AI model at all. So no AI systems learn directly from session content. Those de-identified and decoupled transcripts are used to evaluate and improve the instructions that guide how our products generate those drafts. And that I would say that’s firmly in the what I called that prompt refinement category, and not a model training category. So I do think that distinction matters a lot, and I wish more companies were a little bit more clear about that. I hope that, does that clarify things a little bit more?

Curt Widhalm 33:34
So maybe to try to reflect back what I’m hearing is SimplePractice will take a a piece of data with client data de-identified. They will use that to test it. They will try and break their model. They will go and make sure that this session with all of this data removed creates the note that it’s actually supposed to and does it accurately. And if it doesn’t, SimplePractice uses that as feedback to improve the AI spread.

Ali Hartley 34:04
We yes yes, so we use that sort of quality assurance feedback to improve the prompts that we’re giving the AI. That then is what generates that note.

… 34:14
(Advertisement Break)

Katie Vernoy 34:15
I think the the big distinction for me and the way that you’re describing it helps me feel very reassured. Is the any clinician that’s using AI at all on their own understands when you try to get an AI large learning model or large language model to write a newsletter, you’re going to to give it information around you know your writing style, those kinds of things, and then it’s going. Then you’re going to play around with the prompts of how does my newsletter actually come out. So you’re not creating a new Claude.

Ali Hartley 34:52
Right.

Katie Vernoy 34:53
You’re actually just teaching your version of Claude how to do the work that you’re wanting it to do, and so it comes out and it an AI sounding really clunky newsletter, and you’re like, nope, that’s not right. And so you say, okay, I’m going to give you another prompt to fix it, and then I’m going to give you another prompt to fix it. You’re not necessarily downloading your whole brain into Claude. You’re also not creating a new Claude. You have your own instance. So for SimplePractice, my understanding is that you have, actually, Claude, you have a a consumer grade or or or what’s the what’s the term?

Ali Hartley 35:26
Enterprise for that.

Katie Vernoy 35:34
An enterprise version of a commonly known large language model that is yours. So, what does that look like? How do we know that that it’s safe?

Ali Hartley 35:44
Yeah. So, so it’s it’s what we call bedrocked within our system, and what that means is completely walled off. We do not send data back and forth to Anthropic. Like it all stays within our technical ecosystem because it is walled off and and lives within SimplePractice, so I think that’s a really important feature. It’s a you know thing because you know people who do use you know maybe other AI tools that are built around those type of like large language models, which which most AI tools are. It’s unclear sometimes whether you’re actually sending data back and forth to Anthropic or whether it is staying within a bedrocked system. If if it’s not bedrocked, you know, my recommendation is to see like, does that vendor have a zero retention policy with, you know, Anthropic or OpenAI or whoever they’re using? Because that is really important. If they are sending data back and forth, it’s important that you know the underlying large language model is not retaining that data in any way.

Katie Vernoy 36:44
So, SimplePractice has their BAA’s. They have their security protocols. They have all those things, and that’s something that everyone want to would want to make sure to check.

Ali Hartley 36:54
Yeah.

Curt Widhalm 36:55
And people have the option to opt in and out of this, right?

Ali Hartley 36:59
Absolutely, absolutely. So let me let me, if I can, walk you through sort of like how SimplePractice handles transcript retention, data de-identification, and all that stuff. Because like, you know, I want to make sure we’re walking through exactly how we do it, if if that’s okay. So the audio, so the I kind of went through all the various steps and where the vulnerabilities were earlier in the in this podcast, and so let’s go back to that and then like talk about you know what we’re doing as sort of like the SimplePractice world. So with the audio retention or you know audio piece, that audio session is deleted immediately once the transcript is generated. So we do not retain any audio recordings whatsoever. The transcript retention, as you guys know, that back starting back around mid June, we did begin retaining de-identified and decoupled transcripts from our Notetaker sessions. And so what we did is, you know, I mentioned we follow that safe harbor method. We remove all 18 categories of those identifiers, and then we take an additional step of decoupling. And I know I’ve mentioned this a couple times, but it’s worth going into a little bit more detail here because it’s beyond truly what what actually the law requires. So decoupling means we sever that link between the de-identified text and any account metadata. So even internally, there’s no pathway from a de-identified transcript back to a specific clinician account or client record. The text exists in isolation; it’s disconnected from the who, the when, the where, the why. And it’s not just that we’ve removed the identifiers from the text; we’ve removed the relationship between the text in any identifiable record in our system. So one of the really important sort of practical consequences of this is that because of that decoupling, retained transcripts cannot be subpoenaed. So what that means is like if SimplePractice gets a subpoena and says, “Give me all” you know you know from a government entity says “give me all the records associated with the X Y Z clinician.” Like we cannot go. We actually there’s no way for us to go and identify and locate or produce a transcript associated with a specific individual clinician or or client. Going then into the the specific opt out mechanics that that you asked about, Curt, which is we believe that clinicians and clients should have full control around whether and how data should be, you know, de-identified and decoupled transcripts should be used. So we built this as a transparent choice, not buried in the setting. And honestly, like this is not necessarily a legal decision, but but really a values one, and it’s made our engineering teams’ life harder. They probably hate us for that, but but honestly, like you know, our view is as a mission-driven company, you know, we don’t get to cut corners on on trust here. So as I mentioned, you know, mid June we made that decision to start retaining de-identified and decoupled transcripts, but existing you know people who are using our Notetaker, prior to that date, were opted out of transcript retention automatically by default, so there was no action required, and no transcripts were retained unless they choose to go back and opt in. Customers who started using our Notetaker product after that date are opted in by default, but can opt out at any time at all. And there are three different levels that they can opt out at, which I think is is really important. The first is what we call sort of a global opt out. So it means like you as a clinician decide zero, you know, transcript, you don’t want any transcript de-identified and decoupled transcripts retained. You can go in and you can you can select that. The second is at the client level, and so as as you as a therapist are having, you know, conversations with your your customer or your clients, and a client says they don’t want anything retained, like you can for that specific client opt out. And then the third is at the session level. So if you’re having a session, you’re having a you know session with a client, and and you’re feeling at the end of that session like, oh man, that was that was heavy and a lot. Like I don’t want this specific session retained. Like you can click it. You can opt out at that session level as well. So those who have chosen to opt out, that transcript is permanently deleted after the note is signed and locked, or seven days, whichever comes first.

Curt Widhalm 41:17
The metaphor that I’m trying to to work with as far as decoupling and de-identifying: is de-identifying is going into a coffee shop and getting handed a coffee without a name written on it, and decoupling is then taking the contents of that coffee cup and dumping it into an ocean of all coffee that has been there forever. Am I understanding that correctly?

Ali Hartley 41:39
I think that’s a pretty decent analogy, yeah. I mean, because you can’t go back and find that coffee and say that was Curt’s coffee. Like, I’m gonna like take a little cut back in here and be like, this, this, this was Curt’s for sure. Yeah, yeah.

Katie Vernoy 41:53
So once the de-identified, decoupled transcript is put into the ocean of coffee, how is it used? What does it What does that look like? It’s is it a clinician looking at it? Did we get the answer to that? Do we know?

Ali Hartley 42:06
Yeah. So it it is actually both. So it’s both AI and people, including clinicians on staff that we have here. So you know we have Lindsay. You guys have met Lindsay, and and she’s building out a team, which is super exciting. It is both AI looking at it and and clinicians, but you know the AI that we are using to look at those transcripts is also you know it is bedrocked within our system. So no data again is going back and forth between us and a external provider.

… 42:37
(Advertisement Break)

Curt Widhalm 42:37
Earlier in the episode, we talked a little bit about the Chrome extensions or things like that. There, and you’ve mentioned several times about bedrock infrastructure and a lot of this stuff. Even as much as I am doing my absolute best to keep my focus on this, is some of the technical stuff that just might make my eyes glaze over and mind start wandering. So, I want to be able to talk a little bit more about some of the technology stuff in ways that make practical sense for our audience because we didn’t go to get computer science degrees.

Katie Vernoy 43:15
Or not all of us did.

Curt Widhalm 43:17
Right.

Ali Hartley 43:17
Nor did I. Nor did I. Yes, we we did talk a little bit about sort of like the native EHR integration, the Chrome overlays, and and sort of browser extensions. And I think there’s a third area which is worth mentioning, which is like manual copy and pasting. Which is like this is an example where like a you know a therapist finishes a session, opens a separate browser or a tab, you know, or a separate application, and paste like raw session notes into a third-party AI tool, right? So I think there are a couple vulnerabilities here that are worth thinking about. The first is, you know, what I’ll call clipboard vulnerability, which is like when you copy and paste onto your clipboard, that data can be accessed by other applications on your device, so some mobile and desktop apps have been caught reading clipboard contents without user knowledge. So if you’re copying PHI into your clipboard, like you’ve created sort of a little bit of an uncontrolled exposure. The other is sort of unencrypted transit. So when you paste into a web app, you’re transmitting that data into the app’s servers, and so if those, if that connection isn’t properly encrypted, or if the app is logging inputs, your client session content could be sitting on an infrastructure you haven’t necessarily vetted. And then I would say the third is there’s really no audit trail there. So your EHR has audit logs of who access what and when, and a copy and paste workflow creates no record, right? So if there is a breach investigation, you can’t really demonstrate where that data went. So thinking about sort of all three of those-the native EHR integration, the the web browsers, and the copying and pasting-I think an interesting way to think about this is like to compare it to a home security system. So, a native EHR integration is like having your alarm, your cameras, your locks all on one platform that’s monitored 24/7. A Chrome extension is like giving a contractor a key to your house and then trusting them to go only into the room that you’ve told them they can go into. And then copy and paste is like photocopying your medical records and handing to them someone on the street with a smile, a promise, I would say. So I don’t know if that helps sort of like demystify that a little bit from the technology piece.

Curt Widhalm 45:30
And I think that some of this ties back into some of the HIPAA parameters that we were talking about, is having your device that you’re doing this on encrypted in the first place as one of the security at rest kinds of pieces, that you know you mentioned having it you know encrypted as it transits. That’s the the security and transit piece of this. But there’s a lot of ways that we end up violating HIPAA without even thinking about how easy it is, and in fact, I think I’m remembering that there was a article in about 2020 where Health and Human Services said basically nobody is able to follow all of HIPAA. We can post that in our show notes over at mtsgpodcast.com But there’s a lot of vulnerabilities here, so if somebody is using multiple web browsers or multiple platforms, what are they supposed to do? Just copy everything over by hand, type it into the other thing? I mean, if copying and pasting is going to violate things at some point, I guess I wouldn’t bother.

Ali Hartley 46:39
It may violate things. It’s the problem is you’re right. Is like we don’t necessarily know, and that and that’s the scary part. So I I guess I’d say like there there are a few different things I would want be wanting to think about. You know, if I were a therapist using multiple tools, there are a few things that I would consider to be non-negotiables. We talked about this, but the first is like a BAA with every vendor, right? Every single one, like including the ones that seem smaller harmless. And I know this sounds a little bit like a broken record, but um, but like BAA.

Curt Widhalm 47:10
But you’re you’re an attorney, and you’re supposed to tell all of us, I mean, I when I do law and ethics workshops, I tell everybody at the beginning: if you’re not leaving this a little bit anxious, then you’re probably not doing this in the law and ethics sort of way. So you are supposed to be telling us this over and over.

Ali Hartley 47:26
Yes, yes. So with every single vendor you’re using, get a BAA, right? Like even the one, like I said, even the ones that maybe seem harmless. And then I would say the second recommendation I would I would make is to map out your data flow end to end. So, like, even if that means drawing it out, like, where does your data enter a tool? Where is it? Where is it stored? Does the vendor use sub processors? Do those sub processors have BAAs? And I can tell you, those answers are yes for all of SimplePractice. But it’s like it’s a supply chain question, right? And it matters because as the covered entity, you know the the therapist, you are responsible for the entire chain, not just the first link, right? So, you know, obviously, SimplePractice, we we publish clear documentation of like how the notes are generated, our transfer retention policies, you know, how our AI processing works, and and we make that available publicly through our trust center, and you know, we have an AI addendum to our terms of service, and so I, you know, just work with vendors to make sure that they do the same. Even if it’s not public, they they may be able to give it to you sort of on a one-on-one basis. The third would be minimize what you’re sharing. So if you are using a separate AI tool, you know even if you’re copying and pasting, maybe take the extra second to clip out client names, dates of birth, you know, other direct identifiers before pasting it in anything, and and this is what we call sort of the minimum necessary principle under HIPAA, which is like you are only supposed to share the PHI that is minimally necessary to achieve the thing you’re you’re sharing it, the purpose you’re sharing it for, and so you know, from practical standpoint, if you’re using multiple tools, like the less PHI you expose, the smaller your risk footprint, right? You know, verify encryption encryption standards with with tools that you’re using. So every tool you use should encrypt data and transit and at rest. You know, like SimplePractice does. You can’t find this information on their website or their BAA. You know, ask them. We talked about this, but we actually use bank level encryption for for all the encryption that we do, and we have redundant data centers. We have automatic encrypted backups, and we, like I mentioned, we have real time around the clock monitoring for security threats. You know, I’d also just find out what sort of like if there are any, you know, we talked about this. We, SimplePractice uses Claude/Anthropic, but like you know, figure out what sort of tool like what bent what tools the wrappers are built around. Most of the AI tools are wrappers built around one of these large language models. Really figure out like does that, we mentioned this before, but is that data going back and forth? If it is, do they have a zero data retention policy, or is it bedrocked within the system? Those are just helpful things to know.

Katie Vernoy 50:10
The, what I’m hearing is that there there is some work that clinicians need to do on their end. We need to make sure that our laptop that we’re using, or what the laptop we’re using, or whatever computer that we’re using, is encrypted. That we’ve taken care of that piece. It sounds like when we’re opening our EHR or our session, our AI scribe, or whatever’s going to be recording or or holding PHI, that we need to make sure that they’re not other Chrome extensions that might be grabbing data, whether it’s Grammarly or or some other thing, and that if we’re using something separate, like a a different AI scribe that’s overlaid onto our electronic health record, that there is obviously a BAA and all of the other pieces that there’s, that they’re using appropriate protocols that the the wrapper that you know the the large language model does has a zero retention policy or no data shared, and so hopefully that information is somewhere that can be found. Am I missing anything big on on the responsibilities that I have as a clinician? I know, Curt, you scared me one point talking about internet access and Wi-Fi. But before we get too crazy with with how much we’re trying to cover in this episode, what what is a potentially a checklist to make sure that I’m staying HIPAA compliant to the best of my ability.

Ali Hartley 51:41
I can go first, and Curt definitely jump in. But you know, you covered a lot of it, and I would just say treat the BAA as a compliance floor. Do they have a BAA? Great, yes, but do they have independent security certifications? Are they high trust certified? You know, are there any publicly reported breaches? So I think that that’s an important thing. We talked about the encryption and the zero data retention transparency, you know, honestly, I think is really important. If the vendor can clearly explain what’s happening, where data goes, how it’s used, I think that’s a really important thing. If they’re, you know, how clear is their privacy policy, and and one piece we haven’t talked about that I actually think is really important and worth mentioning is clinical alignment. Like was a tool built for therapists specifically, or is a general purpose tool that’s being marketed to healthcare? And I think this actually matters more than people think because a tool designed for behavioral health like understands the unique sensitivity of mental health data, understands that therapeutic relationship, understands those ethical obligations of confidentiality that honestly go beyond HIPAA, but an AI trained on orthopedic surgery or on how to help you write better, Grammarly. I’m not trying to dis Gramally by any stretch of the imagination, but it’s going to solve different problems, right? So, like, you don’t want your progress note to read an MRI interpretation or you know, an article you’re writing for the Wall Street Journal. So I think like it’s important to think about that clinical alignment. And then I would also add: Is there clinician oversight and control? So does the tool let you review and edit everything and anything before it’s finalized? I think that is a really critical piece because the principle here is that the clinician should always be the decision maker, not the AI, right? So, and it actually this aligns with where regulators are heading to. You know, there’s multiple states now pass laws requiring that AI in healthcare assist, not replace human clinical judgment. And so, you know, what is that tool trying to do there? I think is a really important piece. So, Curt, I’ll let you weigh in as well.

Curt Widhalm 53:43
I like what you’re saying because it is the emphasis that Katie and I hold as one of our principles. It’s one of the things that you mentioned earlier about SimplePractice is built for helping clinicians do their jobs easier. It’s not to you know offload all of our work to AI. Really, in hearing you describe this, it’s kind of like dating. We have our basic principles that we want things to meet. The BAA is kind of the here is the basics that I want for somebody as a dating partner, and then all of these other things are what makes a really good relationship on top of that. We can settle for the bare minimum, and there might be people in our field who are happy or even unknowingly dating below where their league is supposed to be. But what we’re really looking for is the more of these tools that we use as clinicians, the more that we are thoughtful about things. Not only are smarter when it comes to actually protecting client data. Two, in case something goes wrong, the more that you as the clinician are protected because you’ve thought of all of these things. And when I talk to clinicians, whether they’re licensed, we just recently did our annual HIPAA training. And went through all of our compliance checks in our practice, and you know, there’s always, oh, we didn’t know that adding this tool in added, you know, another potential thing that we want to make sure that everybody is on par for. These are things that it’s an ongoing work, and for all of our relationship therapists out there, relationships are ongoing work, and that includes with the tools that we’re doing, that we’re dealing with.

Ali Hartley 55:25
Agreed. Yeah.

Curt Widhalm 55:28
One thing I haven’t heard you talk about is making sure that the data centers are also based in the country that you are working in, because laws about access to that data can change based on the jurisdiction of where the servers are.

Ali Hartley 55:44
You’re absolutely right. It’s a great point to bring up, and it’s an important one for clinicians to think about because oftentimes any payer contracts they have, they’re taking Medicare, Medicaid, will have requirements in those contracts to that you know PHI cannot be stored offshore, so that is a really important piece. And yeah, SimplePractice stores our PHI all all onshore.

Katie Vernoy 56:09
We’ve talked a lot about how someone can address their HIPAA requirements in the current landscape, how they can protect their clients’ data, how they can make sure that the tools they’re using are taking due care of of those responsibilities as well. What are some emerging risks? What what do we need to be paying attention to as we go forward? Because AI is just becoming more and more ubiquitous, I guess. It’s it’s it’s involved in so many different pieces of practice at this point, and as the the use becomes more clinical, you know, treatment plans, those types of things, or even takes on some client face to face type work, where you know there’s a an AI bot that’s making sure that your client walks through the the intake process, for example, I think there’s there’s a lot that I think we want to be aware of, and so what should we what what should we be paying attention to, and how can we prepare ourselves for the the future?

Ali Hartley 57:14
Yeah, it’s a great question, and I would say in addition to things you just mentioned, everyone wants a piece of AI right now, and there’s like so much money out there funding new startups, funding new companies that everyone wants a piece of it, right? And so, I think there’s a there’s a few things that I would say just pay attention to and and be aware of. The the first is you know what I’ll call vendor sprawl, right? So like as as we have more and more AI tools, it’s tempting to adopt, you know, three or four different ones for different parts of your workflow. And I get it, right? Like, this one may do one thing really, really well, and this one may do another thing really, really well, and maybe that is the right choice. But every new ben a new tool you’re adopting is a new data relationship when we talked about those relationships, right? So it’s a new BAA, a new potential breach vector. So the more tool tools you use, the more complex your compliance footprint becomes, and and the harder it is to track where PHI is going, honestly, so like I would just say like one, be aware of vendor sprawl, be aware of the free tool trap. Honestly, like if an AI tool is free and there’s no BAA, ask yourself how is this company making data, and the answer or making money, and the answer may be your data. Honestly, so like free tools aimed at healthcare providers that do not offer BAA’s are everywhere right now, and I actually think they represent a really big unrecognized risk in private practice because it’s just not clear what’s happening and what they’re doing. Right. Another sort of big emerging area right now is is regulations. The compliance current landscape is shifting really, really fast. You know, health Department of Health and Human Services late last year issued a major request for information specifically focused on AI clinical in clinical care. The proposed HIPAA security rule updates, you know, tightened cybersecurity requirements. Multiple, as I mentioned, multiple states are passing AI in healthcare laws, and so there is, and and on top of like that that sort of what I’ll call a 50 state patchwork, which is like really difficult to keep up with, particularly if you’re practicing across multiple state lines. But there’s also this active tension happening at the federal level over whether to preempt state AI laws or let that patchwork continue. So practitioners who assume my EHR handles my compliance for me, may find themselves caught off guard. You know, when a new obligation kicks in, that could apply to them as a covered entity and not just to their vendor. So one of the things SimplePractice is doing to try to help with this is really, you know, we are very much on top of all these laws that are changing and coming out there, and we have a regulatory monitoring we’re doing on an active basis, and so you know we’re trying to make sure we’re staying ahead of all this and looking around all the corners. And also, we want to help you be aware of all these new laws that are coming out too, even if you know it does apply to the to the practitioner and not necessarily the vendor. But you know, we we sort of view ourselves are as as your administrative partner, right? So we’re going to be doing a lot of work in that area. I would say another sort of risk area for clinicians is liability for AI-generated content. Truly, so you know when an AI generates a clinical note and you sign it, you own it. So if that note contains an error or an inaccurate clinical impression or you know, mischaracterized statement from your client, like you are professionally and legally responsible for that. So AI generated notes do require the same level clinical review as notes that you wrote yourself. So the AI tool, you should view the AI tool as helping you save time on the drafting, but it doesn’t save you from the responsibility of of reviewing of what goes into that medical record. So yeah, I think those are some some areas I’d be thinking about and looking out for.

Curt Widhalm 1:01:09
And I think that this is less of a technical issue or or even a legal issue, but with all of these tools that are available, and I’m glad that you’re bringing up the responsibility that goes when you sign a note is you’re ultimately the one who’s responsible. And with all of these tools and artificial intelligence that’s there, it can be really tempting to just become very dependent on them. But we know that artificial intelligence can make things up. It can hallucinate. It can mis-identify things. It’s an ongoing developing tool. It’s why SimplePractice is using it to better the tools and using those transcripts to make it more accurate. But there are always going to be quirks, and when you do sign off on things, you’re putting your licensure on the line by saying, “I have reviewed this, and everything is accurate here.” But there is that temptation to become just oh, it’s been right enough most of the times. I can just get in the habit of clicking yes, I, this is what happened.

Ali Hartley 1:02:11
Yeah, agreed 100% It is, and and you know we’re not just seeing it in healthcare, but you know across everyone’s seeing that with the use of AI now, as people are getting so reliant on it that you know you might be having a convers Slack conversation with a colleague, and they’re like, “Man, that reads like AI,” you know, so, so like you know, we just like as humans, just really need to be conscious of like you know how how much of this is like it just really reflect who I am, you know.

Katie Vernoy 1:02:40
We have a couple of episodes I want to mention. They’re both actually continuing education. One is ways in which supervisors can make sure that their clinicians are not losing skills or not are failing to gain skills or not failing to gain. You know what I mean. We want to make sure that…

Curt Widhalm 1:02:57
Clinician deskilling.

Katie Vernoy 1:02:58
Yes, deskilling is such, I hadn’t heard that word before you brought it up, Curt. So that’s why I was trying to describe it in a different way. But we have a CE on how to make sure that we don’t lose our skills or or fail to gain skills because of AI use. And then we also have a more recent episode, and and that one will be I’m sure become dated as all our AI episodes do. But one where we talk through the AI ethics codes that are coming down to make sure that that as we continue to say even if AI created it you own it and you’re responsible for it so we’ll link to both of those in the show notes over at mtsgpodcast.com.

Ali Hartley 1:03:36
Yeah the the ethics piece Katie you bring up is a really great point because you know you don’t only only own like that legal responsibility for the medical record, which is like where my lawyer brain went. But you also, you know, I’m not a clinician, so but but you guys already like you own that trust with your client and you own that relationship also. And so you know, being transparent about how you’re using AI in your practice with your clients, helping them understand all that stuff is a really important piece. Like the consent shouldn’t be treated as you know a check the box, you know, type of thing. Like have those conversations. It’s it’s a really important thing that to maintain trust with your clients from an ethical standpoint.

Katie Vernoy 1:04:15
Practically in using AI for my practice, I find that there are some real benefits in that I’m able to retain more information in my brain because I’ve got a listening partner, and it makes it easier to get stuff done, honestly, but there’s also that piece of of making sure that I’m reviewing it, that I’m not defaulting to what’s there, and that I’m having those conversations with my clients about what’s going on. It truly is, I think, the future. I think a lot of folks, and and we’ve talked about this in the past, but I think that that a lot of folks are concerned about AI, and I think with good reason. And it’s something where the more that we can understand it and use it properly, and make sure that we’re using it safely, the stronger I think we’ll be able to sustain an independent practice in the current landscape. That’s a whole other conversation, and we’ve actually had it before. But I think there’s this element of if we know that we’re taking care of our responsibilities, at least for me, it it creates a sense of security, a peace of mind.

Ali Hartley 1:05:31
I guess I would just like close by saying, therapists went in this profession to help people, not to become privacy experts or data experts or security experts or lawyers, or CEOs, or whatever it is, and I know that like we talked a lot today, and that like all the amount of information probably feels really overwhelming. But what I would say is like the good news is you don’t have to become those things. Like just be intentional about the tools you use and the partners you trust, right? So ask the hard questions, read the BAA. Something feels good, too good to be true. It probably is. Trust that instinct. So, I would just say, you know, you don’t have to become an expert. Just be intentional. I think that that would be a key takeaway for me.

Curt Widhalm 1:06:15
And you don’t need to fear artificial intelligence. You do need to respect data boundaries, you do need to get your BAA’s in order. You need to stop treating your web browser like a shopping extension that will give you discounts on every other shopping site. But you do need to be intentional about the tools that you’re using and to be able to use the benefits of this, and when done right, it isn’t creating robots-at least here-that are out to replace your jobs,

Ali Hartley 1:06:49
Right. And and with every new technology, we have been through this transformation of like thinking through like what does this mean for HIPAA and PHI and all this stuff. Like you know, when everyone went from you know in place servers to cloud storage, like that was a this was a conversation back then too. So every new technology has this moment in time where we’re like, oh my gosh, what does this mean for for data security and all that stuff? But like you’re absolutely right, doesn’t mean that that fear and avoidance is the answer. It just means like understanding how how to apply that to your practice.

Curt Widhalm 1:07:26
So big thanks to Ali Hartley for pulling back the curtain on how we operate with tech, and especially her knowledge about how SimplePractice in general handles the security behind the scenes, so we can worry less and get our notes done in time for lunch and all of the other things. So, Ali, if people have more questions about this, is there a great way to be able to contact SimplePractice and follow up with more?

Ali Hartley 1:08:00
Yeah, so I I would say like we’ve got a like that our Trust Center is fairly new, so it’s a really great place to start. So if you if you just search SimplePractice Trust Center, you should be able to find it. But also like I’m willing to answer any questions, and I and I welcome the questions and I welcome the scrutiny truly because like we want to be as transparent as we can be, and so people will like, I’m happy if anyone wants to email me directly and ask questions. My email address is ali.hartley@simplepractice.com, and Ali is just spelled A-L-I. So totally happy to take any direct questions. Also, you can email questions to our privacy inbox at privacy@simplepractice.com. So lots of different ways to get a hold of us for questions about this type of stuff, and we’re happy to happy to help.

Curt Widhalm 1:08:46
And we will include that in our show notes over at mtsgpodcast.com. For those getting CE’s, head over to moderntherapistcommunity.com, or if you don’t remember that, go to the show notes too, or listen to the beginning and ending of the episode. Follow us on our social media. Join our Facebook group, the Modern Therapist Group, to continue on with this and other conversations. And until next time, I’m Curt Widhalm with Katie Vernoy and Ali Hartley.

… 1:09:12
(Advertisement Break)

Katie Vernoy 1:09:13
Just a quick reminder: if you’d like one unit of continue education for listening to this episode, go to moderntherapistcommunity.com, purchase this course and pass the post test. A CE certificate will appear in your profile once you’ve successfully completed the steps.

Curt Widhalm 1:09:28
Once again, that’s moderntherapistcommunity.com.

Announcer 1:09:32
Thank you for listening to the Modern Therapist Survival Guide. Learn more about who we are and what we do at mtsgpodcast.com. You can also join us on Facebook and Twitter, and please don’t forget to subscribe so you don’t miss any of our episodes.

 

/0 Comments/by
0 replies
SPEAK YOUR MIND

Leave a Reply

Your email address will not be published. Required fields are marked *