WHITE PAPER
Security, Privacy, and AI Safety
How KindredMind protects a vulnerable person, a family caregiver, and the family's voice.
Contact us for access →

Every call KindredMind handles follows the same principles that trained dementia care professionals are taught on day one: warmth before information, patience before resolution, comfort before accuracy.

The Alzheimer Society of Canada's published guidance for dementia-friendly phone calls, the accepted standard in professional memory care, doesn't ask caregivers to be literally accurate. It asks them to be warm, patient, and reassuring. To meet the person where they are.

That is the foundation every KindredMind call is built on.

Protecting your loved one, and your family's privacy, means building on that foundation with the same rigor. Here is exactly what we do.


Data Minimization

The strongest privacy protection is the data we never collect. KindredMind is built so families never need to share medical records, diagnoses, or sensitive personal details. We do not collect your loved one's full legal name or date of birth, and we ask families not to share medical, financial, or other sensitive information. What we do hold, the voice model and the everyday details that make a call feel familiar, is kept to a minimum, encrypted, and can be deleted at any time.


Consent and Legal Authorization

The service is set up and authorized by the family caregiver responsible for their loved one's care, ideally the person who holds legal decision-making authority. This is not a technicality. It is the framework that governs every care decision made on behalf of a person who lives with dementia: which facility, which medications, which care approaches, and which communication tools serve their wellbeing.

The person with dementia is not informed that calls are being answered by AI. This is a deliberate design decision grounded in the established ethics of dementia care, which prioritizes the emotional wellbeing of the person with dementia over factual transparency. The same clinical framework that guides professional dementia care communication supports this approach. We have written about this in detail.

The companion's voice presence is built only from your own short interview sessions with Sarah, our AI-guided voice setup, completed during onboarding, or, if you choose the assistant voice, from no voice recordings at all. Real conversations between the companion and your loved one are never used for training. Anything you want the companion to learn from a real call, you add yourself by opening the Voice tab and starting another short interview session.

Before we build your voice model, we ask for your explicit consent to use your recordings for that single purpose. Your voice is yours: it is only ever used to build your own companion. Never to build anyone else's. We never sell it. You can delete it at any time. The recordings themselves are automatically deleted within 24 hours of your voice model being ready. Read the full voice model guide →

You never have to record your own voice.

If recording your voice does not feel right for you, choose the assistant voice instead. It is a warm, natural companion voice that works exactly the same way, answering calls and making gentle check-in calls and reminders, and it draws only from the knowledge base you build about your loved one.

For a full treatment of our ethics and consent framework, read our ethics position. Read our ethics position →

Consent and authority when KindredMind reaches out

KindredMind can now make calls as well as answer them: scheduled check-in calls and gentle reminders in your own voice or the assistant voice. Because a call we place is more active than a call we answer, we hold it to the same protections, plus a clear standard of authority.

You confirm your authority to authorize it. Before you can turn on reaching out, you confirm that you hold the legal authority to authorize calls to your loved one, for example as their power of attorney or substitute decision-maker. We record that confirmation, with the date, as the basis for these calls. We do not independently verify power of attorney. We rely on your confirmation, and you remain responsible for having the authority you attest to.

Your loved one's wishes come first. If your loved one asks the calls to ease off, we honor it right away, even though you authorized them. Their own wishes take priority over the setup, always.

Reaching out is off until you turn it on. You choose whether to enable it, which days and times, and how often. Calls happen only inside safe daytime hours in your loved one's timezone, never late at night. If a call is not answered, we try once more and then leave it, with no repeated ringing.

The same safety applies. Mood monitoring, distress alerts, and post-call summaries work exactly as they do on answered calls, and KindredMind never tells your loved one you are on your way when you are not.


Data Privacy: What We Collect and Why

KindredMind collects information in two categories: the information you give us directly (account details, intake data, knowledge base entries, voice recordings), and the information generated by calls (transcripts, timestamps, call metadata).

What we collect:

What we do not collect:

Encryption: What Is Protected and How

KindredMind is built privacy-first from the ground up; we never collect your loved one's full name or date of birth. The personal details you do share, their life story, every word of every call, your phone number, are encrypted before they are ever written to our database. If our database were stolen, an attacker would find nothing but unreadable ciphertext.

Everything sensitive is encrypted at rest using AES-256 encryption, the same standard used by financial institutions and healthcare providers. This includes:

Data in transit between your browser and our servers is encrypted via TLS 1.3. Voice presence files are stored in encrypted object storage with restricted access.

The encryption keys are stored separately from the data they protect. Decryption only occurs at the moment data is needed, for call handling, dashboard display, or summary generation, and only within our secure server environment.

"This is information about your loved one's fears, memories, and daily life. It deserves the same protection as a medical record. That is exactly how we treat it."

No Training on Your Data

KindredMind will never use your conversations, your loved one's words, your voice recordings, your intake information, to train AI models. Not our models. Not third-party models. Not any model.

We are explicit about this with all of our AI providers. Our contracts include data processing agreements that prohibit training on customer data. This is not a preference, it is a contractual requirement we enforce with every vendor in our stack.

Your interview-session recordings, the raw audio used to build your voice, are automatically deleted within 24 hours of your voice model being ready. Your voice model itself can be deleted anytime from your dashboard and is permanently removed within thirty days of your request. Call transcripts are encrypted and automatically deleted after 90 days.

When you delete your data, through the account settings or by closing your account, it is permanently removed from our systems and from the systems of our providers. We do not retain backups of deleted user data beyond the standard 30-day recovery window.

Care Circle Plan

The Care Circle plan lets up to four caregivers share one subscription and one knowledge base. Here is how we handle data within a care circle, and what each member can and cannot access:

All care circle data is subject to the same encryption, retention, and deletion commitments described on this page. The fact that data is shared within a circle does not change how we store or protect it.

Call Safety Protocols

Every call handled by KindredMind goes through a real-time safety layer that runs from the first word to the last. It is not a post-call review. It is built into the call-handling engine itself, analyzing conversation as it happens and acting before the call ends if it needs to.

What triggers an alert

KindredMind monitors every call for two things in parallel: a set of built-in phrases associated with physical emergencies, and a broader AI analysis that catches distress, fear, confusion, and safeguarding concerns that no fixed keyword list could anticipate.

The built-in emergency phrases include things like:

I fell I can't get up can't breathe chest pain heart attack fire call 911 I'm bleeding help help bad pain + more

These phrases always active. You cannot turn them off entirely, but you can mute up to 10 of them from your dashboard settings if a specific phrase is a false alarm for your loved one — for example, if they say "you have my heart" constantly. Muting a phrase removes it from keyword detection only; the broader AI analysis still runs.

You can also add your own custom trigger words from dashboard settings. These are additive: they layer on top of the built-in list, not instead of it. A word you add triggers the same alert sequence as any built-in phrase.

When a keyword is matched, a second AI pass confirms whether it is a real physical emergency happening right now or a benign mention — a memory, an idiom, a story. When in doubt, it treats it as real. Missing a real fall is a worse outcome than a false alarm.

What happens when an emergency is detected

The moment a real emergency is confirmed, five things happen simultaneously:

Backup contact voice cascade

If your primary caregiver number is in quiet hours or the call cannot be placed, KindredMind immediately moves through your backup contacts in the order you have set them, skipping any contact in their own quiet hours, until it reaches someone who can be called. If no one can be reached by phone, the alert is logged and marked unresolved so you can see it in your dashboard. SMS messages to all contacts go out regardless.

Emergency escalation sequence
1
SMS all contacts immediately — primary caregiver + all backup contacts, any time of day
2
Voice call to primary caregiver — skipped if you are in your quiet hours
3
Voice call to backup contact #1 — if primary was skipped or unreachable
4
Continue cascade through remaining contacts — in your set order, skipping anyone in their own quiet hours

Other alert types

Not every alert is an emergency. KindredMind monitors for a range of call conditions and adjusts the notification accordingly:

KindredMind is not a medical monitoring device and makes no claim to replace one. It does not guarantee it will detect every emergency. What it does: listen carefully, act immediately when it hears something serious, and make sure a real person is notified while keeping your loved one company until help arrives.

No false promises

The companion will never tell your loved one you are on your way when you are not, that something will be arranged when it will not, or that a problem has been fixed when it has not. It operates within honest limits. When an emergency is confirmed and the escalation sequence is running, the companion says exactly one thing: help is on the way, stay calm. That is the only promise it makes — and it is one the system is actively working to keep.

Post-call review

Every call generates a full summary: key topics discussed, any flagged moments, the mood reading, and your loved one's language patterns if anything notable was observed. Nothing is hidden. The transcript is there if you want it, and the summary is there if you do not. Anything flagged during the call that did not trigger a real-time alert is highlighted in the summary so you can decide whether it warrants follow-up.

Your Control Over the System

You are in control of KindredMind at all times. There are two modes:

You can switch between modes instantly from your dashboard. If you are having a moment, if you want to take every call yourself today, or if you want to be completely off-grid, the system adapts immediately.

You can also export all of your data at any time from account settings. This includes your intake form, knowledge base, call transcripts, and account information, in a single JSON file. You own this data. You can take it with you.

Our Privacy Commitments in Plain Language

Data

Your data is yours

Your voice is only ever used to build your own companion. Never to build anyone else's. We never sell it. You can export or delete everything at any time.

Security

Encrypted at rest and in transit

AES-256 encryption for all sensitive fields. TLS 1.3 in transit. Voice files in restricted encrypted storage.

Safety

Distress alerts on every call

Real-time distress alerting with immediate caregiver notification. The Emergency Failsafe routes the call to the caregiver's phone if the companion or server fails.

Control

You are always in charge

Two modes, instant switching, full post-call transcripts and summaries, and an automatic failsafe that hands the call to your phone if anything goes wrong.

Honesty

No training on your conversations

Contractually prohibited with all vendors. Your loved one's words never improve models for anyone else.

Compliance

Privacy by design

We follow applicable privacy law in every region we serve, including PIPEDA in Canada and US state privacy laws. Our data practices are built around minimization, not compliance checklists.

Third-Party Providers

KindredMind uses a small set of carefully selected third-party providers. We have data processing agreements with each of them that explicitly prohibit training on customer data.

We do not use advertising networks, analytics that track individuals across sites, or any provider whose business model depends on data monetization.

Reporting a Concern

If you have a privacy concern, a question about your data, or a safety issue with a call, please reach out directly at hello@kindredmind.care. We respond within 24 hours on business days.

If you believe there has been a data breach or a safety incident, contact us immediately at the same address and include "URGENT" in the subject line.

Trustworthy by design.

If you have questions that are not answered here, we want to hear from you before you sign up, not after.

Ask Us Anything

Understanding why this approach is ethical →
What families try before finding KindredMind →

Further Reading

For the full legal details, see our Privacy Policy and Terms of Service. For our ethical framework around AI in dementia care, see Our Approach. For the question of whether AI is appropriate in this context at all, see Is It Ethical to Use AI for Dementia Care?