ESSENIALAI & Memory Transparency
1. What may be generated
Subject to consent and product availability, software may generate a conversational answer from archived sources, a synthetic reading in an approved voice, animation from an approved portrait, a representation at an approved age, a summary, translation, transcript or suggested question.
2. What remains original
Original interview audio/video, uploaded photographs, documents and attributed witness statements remain distinguishable from generated presentation. A visible “Source” control should identify the relevant recording, transcript, date and contributor without exposing material outside the visitor’s permission.
3. Labelling
At first interaction, the experience will say that it is AI-assisted and archive-based. Materially generated or manipulated audio, images and video will carry a clear, accessible disclosure and, where technically and legally required, machine-readable provenance. Users may not strip labels or misrepresent synthetic media as original.
4. Grounding and uncertainty
Answers should be restricted to authorised archive material and attributed human contributions. When evidence is missing, the system should say that it does not know. It may invite the visitor to choose a well-represented life period or, with permission, route an unanswered question to the Life Subject, Steward or an opted-in Witness.
Memories are subjective. A source proves where a statement came from, not that it is objectively correct. Conflicting accounts should be attributed rather than silently merged into false certainty.
5. Consent for voice and likeness
Creating or materially changing a synthetic voice, animated likeness or age representation requires separate, informed, recorded and revocable consent from the Life Subject while capable, unless documented lawful authority clearly permits otherwise. Approved source recordings and images must be identified. Consent to preserve a story is not automatically consent to clone a voice, animate a face, publish an archive, train a model or use material in marketing.
6. Safety boundaries
- No claims of consciousness, survival, supernatural contact or the person’s current opinion.
- No new will, testament, medical instruction, financial direction, legal authorisation or consent attributed to the persona.
- No manipulative pressure, dependency design, concealed advertising or exploitation of grief.
- No deceptive impersonation, fraud, political endorsement, identity verification or biometric surveillance.
- No non-consensual sexual, humiliating, hateful, violent or defamatory synthetic content.
- Age presentation affects appearance only; it must not falsely imply that words were spoken at that age.
7. Human control
Authorised people can review sources, correct or contextualise material, restrict audiences, disable generation, withdraw consent and request export or deletion subject to law. Significant access, moderation and authority disputes should offer human review.
8. Training policy
Draft default: private stories, transcripts, voice, likeness and family testimony are not used to train general-purpose or third-party AI models without a separate, specific opt-in. Service providers must be contractually restricted to providing the selected function. This commitment must be verified against every production AI provider before launch.
9. Emotional wellbeing
People experience grief differently. Visitors should be able to pause, mute synthetic voice, choose text, leave the experience and avoid unexpected activation. ESSENIAL is not mental-health care or an emergency service. If an experience feels distressing, stop using it and consider contacting someone you trust or a qualified professional.
10. Report or challenge an output
To report a false, harmful, unconsented or mislabelled output, use [IN-PRODUCT REPORT CONTROL / TRUST & SAFETY CONTACT]. Include the persona, question, output and reason. We may preserve the report for audit, restrict the output and ask an authorised human to review it.