Responsible AI Statement

Every product decision starts with: could this harm a patient? If a feature could increase the chance of a wrong health decision, we do not ship it — even if it is convenient or impressive.

Medipher is designed to support the relationship between a patient and their clinician, not to replace it. Explanations should help you ask better questions, not skip the appointment.

The model is instructed to never diagnose, recommend treatment, or say whether something is safe or serious. Where the document doesn't say, we say so.

We publish what models we use, what data we process, and what we can and cannot guarantee. When we can't promise something, we say so instead of hiding behind marketing language.

Large language models are trained on data that reflects the biases of that data. We try to reduce impact by keeping outputs source-bounded (explain the document, not infer), avoiding demographic assumptions, and inviting feedback on unclear or unfair wording.

We do not store the text you paste or the AI's response. Waitlist, feedback and contact data are held only as long as needed. Access to production systems is limited to the people who need it.

We treat quality and safety as ongoing work: reviewing prompts, testing new models against safety criteria before rolling them out, and improving how the app handles edge cases such as fragmented text, unusual results or non-English documents.

Every explanation and every guide can be rated. Feedback goes to a real inbox and shapes what we change next. If something is confusing, misleading or wrong, please tell us.

We follow the direction of travel set by the UK AI regulatory principles (safety, transparency, fairness, accountability and contestability, redress). We are a small team and we don't claim external certification — this is a statement of intent and practice, not a compliance badge.