We only record people who agreed to be recorded
MemoriesBox builds a voice only from recordings the person made themselves, knowingly, while alive and taking part. We do not reconstruct anyone from photographs, messages, or another person's description of them.
Facts on this page verified
The short answer
There is one rule underneath everything else here. The person is recorded while they are alive, they know they are being recorded, and they take part on purpose. Nothing is built from a dead person's photographs, letters, text messages, social media, or a relative's description of how they used to talk.
That rule is not a feature. It is the reason the product is shaped the way it is, and it is what makes it different from most of the things this category is criticised for.
What is being consented to, exactly
Consent to be recorded is not the same as consent to be modelled, and neither is the same as consent to be answered from after death. There are four separate permissions, and a service that collects one and assumes the rest is not asking properly.
| The permission | What it covers | How it is given in the product | What is recorded |
|---|---|---|---|
| To be recorded | Audio of each answer is captured and stored | The person presses the record button themselves, for every answer. Nothing records on its own, and nobody can press it for them from another account | The time of every recording |
| To have a voice model built | A model of their voice is made from those recordings | A separate, deliberate act: pressing "Create my digital self" in Settings → Voice. Recording never builds a voice by itself, and many members record and never press it | The time the model was built and how many answers it came from |
| To let named people hear it | Who gets a key | The storyteller adds each person by name and email in Settings → Trusted contacts, and can revoke any of them. Keys open recordings only when the archive is released | The time each key was issued or revoked |
| To let it continue afterwards | Whether heirs may keep copies of the recordings once the archive is released | A switch in Settings → Trusted contacts, off by default. The release itself is performed by us on an heir's request and can be reversed | The setting, and who released the archive and when |
These are four separate actions at four separate moments, each with a timestamp, and none of them is a checkbox in the Terms. What we do not do is put a spoken "do you agree" before every session: the act of pressing record, each time, is the consent for that recording, and it can be withheld by not pressing it.
The Cambridge safeguards, one at a time
In May 2024, Katarzyna Nowaczyk-Basińska and Tomasz Hollanek of the University of Cambridge published a paper in Philosophy & Technology on what they called the digital afterlife industry. It warns about unwanted "digital hauntings", about deadbots advertising products in the voice and style of the deceased, about unsolicited notifications, and above all about simulations built without the data donor's consent. Their proposed safeguards read like a product specification, so we have treated them as one.
| Cambridge safeguard | What MemoriesBox does |
|---|---|
| No advertising through a simulation of the deceased | There is no advertising in the product, and nothing is ever promoted through an avatar. |
| Age restrictions on who can interact | None built. The product is sold to adults, the person recorded has to take part themselves, and the digital self is reachable only by the account holder today. We say "none" rather than describe a gate we have not built. |
| Persistent reminders that this is AI | Every conversation is labelled as AI at the first interaction. Required of us anyway by EU AI Act Article 50 since 2 August 2026. |
| Opt-out protocols that allow emotional closure | Settings → Voice → "Retire my digital self", behind a confirmation: the voice model is deleted at the vendor and the character brief erased, and the recordings stay. Deleting the account, in Settings → Privacy, removes everything. |
| Consent prompts at creation | The voice is built only from the storyteller's own recorded, consenting participation. |
| Dignified retirement | Retirement is deletion, not a pause: the voice model and the brief are destroyed and cannot be restored, though a new one can be built from the recordings later. The storyteller can do it while living. After the archive is released, any heir holding a key may ask us in writing and we will do it; a legal representative may ask at any time. |
The paper's sharpest point is the one about consent, and it is worth stating in the negative. Seance AI builds its simulations from the surviving person's description of the deceased, and says on its own site that it is for entertainment purposes. In Arizona in May 2025, an AI video of Chris Pelkey, a murder victim, delivered his own victim-impact statement in court. Those are reconstructions of someone who cannot be asked. What MemoriesBox holds is a recording of somebody who was asked, answered, and knew what it was for.
Where consent gets hard
The difficult case is not a philosophical one. It is a Tuesday afternoon with a parent whose memory is going.
7.4 million Americans aged 65 and over are living with Alzheimer's in 2026, projected to reach about 13.8 million by 2060 (Alzheimer's Association, 2026 Facts and Figures). A diagnosis is one of the most common reasons people start recording, and it is also the situation where consent is most fragile. Capacity is not binary, it moves through the day, and a person can be delighted to tell a story at eleven in the morning and frightened by a microphone at four.
Our position is that consent to record has to be renewed by the person, in the moment, not collected once by a family member on their behalf. In the product that is literal: every answer starts only when the person presses record, so a session in which they do not want to talk is a session in which nothing is recorded. Practically, it also means keeping sessions short and stopping when someone does not want to continue.
The policy, written down:
- A family member may set up the account and sit with the person — that is what VIP Setup is — but may not record in their place. The recordings are made by the person, in their own voice, pressing their own button. We do not open proxy accounts for an adult who cannot take part.
- We do not re-ask a spoken consent question at the start of each session. The act of pressing record, each time, is the consent for that recording. We would rather that than a ritual that a person with changing capacity learns to tap through.
- Recordings already made stay if capacity is later lost. They were made knowingly, at the time. They are deleted only if the person asks, or if a legal representative — a health and welfare power of attorney, a court-appointed deputy, or the equivalent in your country — asks us in writing.
- A family can ask us to stop. Write to hello@memoriesbox.org. With a legal arrangement in place we retire the digital self, stop the weekly question email, and, if asked, delete the recordings. Without one, we will still retire the digital self on the request of the account holder or, after release, of any heir holding a key.
What we can't promise
- We cannot promise this is good for grief. Mary-Frances O'Connor at the University of Arizona warns that these products risk making the past too attractive to let go, and Robert Neimeyer, in the same Scientific American piece, puts the share of bereaved people with the anxious attachment most at risk of prolonged grief at 7 to 10%. We are not a treatment and we make no therapeutic claim.
- We cannot promise the business model and your interest always point the same way. Psychology Today put it plainly in July 2026: "The healing process and the business model point in opposite directions." A subscription profits from engagement. Our answer is to remove the engagement mechanics, not to deny the tension.
- We cannot promise the evidence is settled. The most cited study in the other direction, by Xygkou and colleagues at the University of Kent (ACM, 2023), had ten participants, who rated digital ghosts highly and became more able to socialise normally. Ten. The evidence is thin in both directions, and anyone telling you otherwise is selling something.
- We cannot verify capacity. We are a recording service, not a clinician. If a family disagrees about whether someone should be recorded, we cannot settle that for you.
- We cannot promise a family will agree about the archive later. Consent given by one person does not bind the feelings of everyone who survives them.
What the law says
- GDPR Article 6(1)(a) makes consent a lawful basis for processing. Article 7 sets the conditions: it must be freely given, demonstrable, and as easy to withdraw as to give.
- GDPR Article 7(3) gives the right to withdraw consent at any time. Withdrawal does not undo processing that was lawful before it, which is exactly why the retirement policy above matters.
- GDPR Article 9(2)(a) requires explicit consent for special-category data. Where a voice is processed in a way that uniquely identifies a person, that higher standard applies. We ask at the higher standard regardless. See /trust/is-voice-cloning-legal-in-europe.
- GDPR Recital 27 states the Regulation does not apply to the personal data of deceased people, though member states may make their own rules. Several have. See /trust/who-owns-a-voice-clone.
- EU AI Act Article 50 requires that people are told they are interacting with AI, and that synthetic audio is marked as artificially generated. In force since 2 August 2026.
Primary texts: Regulation (EU) 2016/679 and EU AI Act Article 50.
Questions people ask
Can I make an avatar of someone who has already died?
Not unless they recorded with MemoriesBox while they were alive. If there are no recordings the person made themselves, there is nothing for us to build from, and we will not construct a voice from photographs, messages or your description of them.
Can I record my parent without telling them?
No. The person has to know, and has to take part. A secret recording is not a life story, it is surveillance, and it fails both our rule and the consent conditions in GDPR Article 7.
What if my mother has dementia?
Record while she can still take part, keep sessions short, and ask again each time rather than relying on a permission given months ago. If she does not want to continue on a given day, that is the answer for that day.
Can the person change their mind?
Yes, at any time, and it is no harder than giving permission was: stop pressing record, retire the digital self in Settings → Voice, or delete the account in Settings → Privacy. GDPR Article 7(3) requires that withdrawal be as easy as consent, and each of those is one screen.
Does the avatar say things the person never said?
It answers from the person's own recorded words and is told not to invent hard facts; when something was never recorded it says, in character, that it does not remember. It is a language model speaking in their manner, so a phrasing can be new even when the substance is theirs, and every conversation says so on screen. How that works is on /features/talking-avatar.
Related: Who owns a cloned voice · What happens if we shut down · Every other way to keep a voice · The full comparison table
If you want to start while the person can still take part, begin here. $139 a year, first 14 days free, cancel anytime.