AI transparency & safety

What exists today

Today YourSomeOne™ is a pre-launch website with an early-access list. There is no companion you can talk to, no account, and no conversation data. Nothing described under “principles for launch” below is implemented yet, and we will update this page as each piece is actually built.

You will be talking to software

When the companion launches, every reply will be generated by artificial intelligence, with no human reading along or replying on the other side. We will never present the companion as a real person, and we will not use fake profiles, fake photographs of people, or claims of human involvement.

What YourSomeOne™ is not

• Not a therapist, counsellor, psychotherapist or mental-health professional.
• Not a doctor, medical device or healthcare service. It does not diagnose or treat.
• Not an emergency, crisis or suicide-prevention service.
• Not a substitute for professional care, or for the people in your life.

If you are in immediate danger or need urgent help, please contact your local emergency services or a qualified professional.

Principles for launch

The rest of this page describes how we intend the companion to work and behave. It is our design intent, not a description of a running product, and not a guarantee about a feature you can rely on today.

How it is intended to work, in plain terms

1. You write; your message is sent to a large language model.
2. Relevant things you have chosen to let the companion remember are intended to be supplied alongside it, so replies stay in context.
3. The model generates a reply. It is prediction, not understanding or belief.

Limits worth knowing

• It can be wrong. AI can state incorrect things confidently. Check anything that matters.
• It has no feelings. Warmth in the wording is a style, not an inner life.
• It can misread you. Tone, sarcasm and cultural nuance are hard.
• It reflects its training. Bias in training data can surface in replies, and we work to reduce it rather than pretend it is solved.

Memory should be yours to control

Continuity is the point of a companion, but it should never feel like surveillance. Memory controls are not built yet. Our commitment is that what the companion remembers will be shown to you in plain language, and that you will be able to edit individual memories, turn memory off, or delete everything, before any of it holds real conversations.

Safety commitments we are designing for

None of these mechanisms are live yet, because there is no companion to apply them to. They are the requirements we are holding ourselves to before launch.

• Adults only; the service is not being designed or intended for children, and we will set out how we handle age at launch.
• Safeguarding responses for conversations that suggest risk of harm, pointing to real-world help rather than trying to handle it alone.
• No manipulative engagement mechanics — no guilt-tripping, no artificial urgency, no designing for dependence.
• Clear reporting route for harmful or inappropriate output.

Model providers

The companion is expected to rely on third-party AI models. We have not settled on our providers, so we cannot name them yet. We will publish who they are and what is sent to them in our privacy notice before launch.

Tell us when we get it wrong

Email hello@YourSomeOne.com. We read every safety report and take them seriously.