Privacy Policy, version 6. Effective 2026-11-05.
This policy covers the exercise app that Mark Bennett LCSW offers online, used in a web browser after signing in with a Google account Mark has approved. It says what the app collects, where that is kept and processed, who can see it, and how it is deleted. It does not cover Mark's other records, which his Notice of Privacy Practices covers, or your own use of Google's services outside this app.
This app is offered by Mark Bennett LCSW, 2775 Channing Way, Idaho Falls, ID 83404 · [email protected] · 208-524-0133. Questions, requests and complaints about your privacy go to that address.
A set of exercises for practice and reflection. A physician can ask someone to exercise; in the same way, Mark can ask someone to do these exercises. They are not designed to replace therapy, any more than physical exercise is designed to replace insulin, and the app is not a crisis service. Mark decides who may use it: people he invites, and people who ask to try it when he agrees. It is designed for adults 18 and over; a child may use it only with a signed consent arranged through Mark's office (208-524-0133).
Because the app is not designed to be therapy, it is not a HIPAA app and is not designed to fall within HIPAA. Even so, it and its AI are built to meet the confidentiality and breach-prevention standards that HIPAA sets, simply because Mark cares about your confidentiality.
The app is built for role-play and practice: you may type things that are not true, or play someone else. Please keep real medical details (diagnoses, medications, test results, insurance numbers) out of what you write, and especially out of anything you send to Mark. The app does not need them to work.
The app runs on Google Cloud, in Mark's own Google Cloud project. That project is covered by a Business Associate Agreement (BAA) with Google. The BAA is a contract, binding on Google under United States law, that requires Google to protect your information, use it only to provide its services to Mark, and tell him of any breach. It applies wherever Google handles the information, inside or outside the United States. Google does not use your information to train its AI.
Not everything happens in the United States:
In every case the BAA applies, and what you save is stored only in the United States.
What you save is kept in that project's private storage, in a space of your own that is encrypted with a key made from your own password. Mark has no copy of that key and cannot open your space; neither can Google or anyone else who sees the storage. Even the names of your folders and files are stored as meaningless codes, so someone looking at the storage cannot tell which exercises you used, what you named anything, or which space is yours. What can be seen is how many spaces exist and how many files each holds, with their sizes and dates. Everything personal in your space (saved sessions, your Parts, Skills and Habit records, notes, screenshots, and the app's own logs) is written through that encryption.
The app does not log your conversations unless you tell it to. Its own technical logs keep no conversation: what you type, the AI's replies and its reasoning are cut out of every log entry before it is written, leaving only what is needed to fix problems (which button, timings, errors, the safety checks' verdicts). The only way a conversation is kept word for word is if you turn on 📋 Logging, which is off until you do; see "What Mark receives."
Each person has a monthly AI allowance (an estimate of what their use costs, $15 a month, the spoken voice included). To keep it, the app records in your own encrypted space one running figure for the month: a dollar estimate and a count of AI requests, nothing about what was said. Past 80% you are told once; at the limit the AI pauses until the 1st of the next month, while everything you saved stays open to you, and the safety checks keep working.
Some exercises (the faith-based ones, and others still being tested) are shown only to people Mark has chosen them for. To do that, Mark keeps a list in the same protected Google Cloud project of the Google addresses he has given them to; it says nothing else, and a person is removed from it when they stop using the app. Anyone can hide any exercise they see from their own list; that choice is kept in their own encrypted space.
While you are signed in, the app's server holds your conversation in its memory so it can work; it is cleared when you sign out, after 20 minutes without activity, or when the server restarts.
If you forget your password, nobody can recover what you saved: not Mark, not Google. You can start again with an empty space under the same name; the old one stays locked.
When you choose a password (when you set up your space, or when you change it) it must be at least 8 characters, must not be one of the most common passwords, and must not have turned up in a known data leak (checked as described below). There are no rules about symbols, numbers or capitals: a short phrase of a few ordinary words works well. A password that is allowed but easy to guess is shown as UNSAFE PASSWORD - PROCEED AT YOUR OWN RISK, and you may use it only after accepting that risk. If a weak password, or one that has appeared in a leak, is guessed or breached, that is your liability; Mark Bennett LCSW does not assume it.
The app can only be reached by people Mark has approved. You sign in first with your own Google account; Google lets you through only if your email address is on Mark's list, and the app receives your email address from Google for that check. Then you sign in to the app itself with your name and your own password, which open your space. Your space is tied to the Google account you set it up with: your name and password do not open it from any other Google account, Mark's included. If you lose that Google account, you cannot get back into the space; you would start a new one. Google keeps a record of sign-ins and of the requests your browser makes (the time, your device's internet address, the page asked for) in Mark's Google Cloud project, without the content of what you write. On a shared computer, sign out of Google when you finish, or use a private window.
The text of the conversation you are having is sent to Google's AI service (Vertex AI, in Mark's project), which writes the replies. The app itself runs on Google's servers, so everything you type reaches Google Cloud. There are no analytics, no telemetry, and no advertising networks.
One service outside Google is asked one question: when you choose a password, the app checks it against the "Have I Been Pwned" collection of passwords found in data leaks. It turns your password into a 40-character scrambled code and sends only the first 5 characters; the service answers with every leaked code that begins with them (hundreds), and the app compares the rest on its own. Your password, and even its full code, never leave the app, and the service cannot tell which code was checked.
If you use the Review page's find-it helper, the text you've queued for Mark is sent to the same Google AI service, once, to run your search. If you switch the Voice button on, the text of a reply is sent to Google's Text-to-Speech service to be spoken, and your recording is sent to Google's Speech-to-Text service and only the words come back; the app stores no audio. Your browser's own speech features are never used, because they send your voice to the browser maker under that company's own terms.
Your saved files themselves are never sent to the AI; what goes is the conversation text needed to write each reply, plus short summaries of your parts, skills, habits and safety plan so the AI can use them.
Google's AI service keeps some of what it is sent for short periods, for the reasons below. None of it is used to train AI (Google's Training Restriction), all of it stays within Mark's project, and all of it is covered by the BAA. This is separate from the app's own safety checks described below.
The app does not use Google's features that store data for longer: it does not use Google Search or Maps grounding, does not turn on request-and-response logging, and does not use Google's stored-conversation service (Google's zero data retention page).
By default, nothing. Mark only ever receives what you choose to send:
The limits of confidentiality. What you send Mark is treated like anything you tell him in person, including the legal limits of confidentiality. Under Idaho's reporting laws, if it contains credible evidence of abuse or neglect of a child, abuse of an elderly or vulnerable adult, or a serious, imminent or foreseeable danger to yourself or someone else, Mark may be required to report it or act to keep someone safe. Other laws may also apply. The app shows you this notice again, together with the therapy-record note below, directly above the button, before you send.
If you also see Mark for therapy. He may decide it is appropriate to add something to your therapy record based on what you send him. A psychology workbook you fill in at home is not copied into that record, and neither is what you do in this app: only what you choose to send, and only what Mark judges belongs there. Anything he adds is kept as that record requires, not deleted with the rest after 90 days.
One small convenience and nothing else: where you last dragged the Feedback window to. That you have read the waiver and this policy is kept in your own space, not the browser, so each person accepts them for themselves. The voice remembers nothing: it starts off every time. There are no tracking identifiers. The app's sign-in cookie holds only a random session number, never your credentials. Google's sign-in also sets its own cookies, which keep you signed in to Google for this app.
The app runs its own automated safety checks on the conversation (crisis language, scope, explicit content). These checks run on the same Google AI service as the conversation itself and can pause or end an exercise and show you crisis resources. The app's own safety checks never notify Mark, the authorities, or anyone else. The app has no way to send anything anywhere except the Google services above and the files you seal yourself.
Google's abuse monitoring is separate, and it is Google's, not the app's. As described under "What Google's AI service keeps, and why," Google may log a prompt its classifiers flag and may contact Mark about it; Google decides under its own terms and the law what else it does with a flagged prompt, which may include reporting it to the authorities. Mark cannot open your space, so he could not look up what you wrote.
If the app's data were ever breached, Mark will tell the people affected without unreasonable delay, and no later than 60 days after the breach is discovered, following the standard HIPAA's breach rule sets. What you saved is encrypted under your own password, which neither Mark nor Google holds.
When this policy changes, the app shows you the new version the next time you open it, and you accept it again, together with the waiver, before you can continue.