Our AI EThics Policy
AI carries the story. It never becomes the storyteller.
Lumenairi® uses artificial intelligence extensively — to help gather a story, to encode it, to share it across languages, and to make it discoverable years later by someone who did not know it existed.
What we do not do, and will not do, is use AI to generate a human being. No synthesised voices. No fabricated expressions. No invented answers. Every word in a Lumenairi® archive was said out loud by a real person who chose to say it.
The Line We Draw
There is a difference between carrying a story and manufacturing one.
The distinction that matters is not whether technology was involved. It is whether the technology changed what the person said, or only how well you can receive it.
Most conversations about AI and video collapse into a single question: is it real, or is it fake? That question is too blunt to be useful, because almost every recording ever made has been edited, lit, levelled and colour-corrected. A microphone is a technology. So is a lens.
Everything Lumenairi does with AI sits on one side of that line. We use it to reduce the distance between a person who spoke and a person who is listening — to clean up the noise in the room, to render the words in another language, to find the two minutes out of nine hours that answer the question you just asked.
We do not use it to put words in anyone’s mouth, to make a face do something the face did not do, or to answer a question the person was never asked.
The technology serves the testimony. It never substitutes for it.
What We Use AI For
Four AI tasks – none of them are the story itself.
Gather
AI builds the question script. It reads the person’s own biography and proposes the questions they are most likely to be asked, so the session is specific to them rather than generic to a category. The interviewer decides what to ask, and can go anywhere the conversation leads. AI proposes; a person chooses.
Encode
Once the recording is made, AI does the work that used to take a production team weeks. Transcription. Translation. Subtitling. Noise reduction and audio levelling. Tagging and metadata. None of it touches what was said — it makes what was said legible, searchable and usable.
Share
AI renders the same recording in other languages, so a story recorded in one language can be received in another without a separate interview. The subtitles and the translated text are AI-produced. The voice and the face are never anything other than the person’s own.
Discover
This is the part that matters most and is least visible. When a viewer asks a question, AI searches the meaning of what the person said and returns the passage where they answered it. It is retrieval, not generation. The answer already existed. AI only found it.
What We Will Never Do
Some things are not on the roadmap, because they are not on the table.
These are not current limitations awaiting a better model. They are commitments.
Lumenairi® will never use AI to:
Generate human content. No AI-written answers, no synthesised speech, no invented recollections.
Create deepfakes or synthetic stories. We will not produce a version of a person saying something they did not say — for any purpose, including ones presented as harmless, educational or commemorative.
Reproduce a person’s physiognomy. We do not generate, model or reconstruct faces.
Reproduce vocal tonality. We do not clone, synthesise or extend anyone’s voice.
Reproduce human expression. We do not alter, generate or animate what a person’s face was doing.
Build or trade in biometrics. We do not create biometric profiles from the recordings in our care.
And one thing we will always do. If a question was not answered, the archive says so. A Lumen will tell a viewer that the person was never asked, rather than offering something approximate in its place.
Why we hold this line
An archive is only worth what it can be trusted to be.
The value of a recorded story rests entirely on one assumption: that the person really said it.
Remove that assumption and everything downstream loses its meaning. A founder’s account of the hard year becomes a marketing asset. A grandmother’s memory becomes a plausible reconstruction. A community’s record of itself becomes a thing that has to be verified rather than a thing that can be relied upon.
The moment a platform is capable of making someone say something they did not say, everything on that platform becomes uncertain — including the parts that are true. Which is why this is not a feature we have chosen to withhold. It is a capability we have chosen not to build.
Archives outlive companies. The people we record will not always be here to confirm or deny what appears under their name. That obligation does not expire, and it is not ours to trade away.
Lumenairi is infrastructure. It carries the story, but it is never the story.
How This is Governed
Written down, reviewed, and auditable.
The original is preserved. Every enhancement is non-destructive. The unaltered master recording is retained and can be produced on request.
Human review. Question sets, translations and published sessions are reviewed by people before release.
Ownership. Your recordings are yours. We do not use content in our care to train third-party AI models.
Improving our own systems. We use anonymised material to improve our own transcription, translation and search — how well we hear a regional accent, how accurately we render a language, how reliably we find the passage that answers a question. This is opt-in. The default is no, the choice belongs to the person recorded rather than to the organisation that commissioned the recording, and it can be withdrawn. No recording is used this way unless the person has said yes.
Named accountability. This policy is owned at founder level and reviewed annually.