Choose which optional cookies and similar storage we may use. Strictly necessary storage is always on, because the site cannot work without it.
Talk to us

Draft: awaiting review by the Chief Technology Officer and the Clinical Safety Officer before publication

Responsible AI and AI transparency

This is our AI transparency statement. It sets out where we use AI, how people oversee it, how we tell you when you are dealing with it, and who is accountable when it gets something wrong.

Last updated

Where we use AI

Prime Assist
Answers and handles patient calls, in several languages.
Averra
In development. Designed to bring a person's health data into one picture and flag changes, pointing them to tests or a doctor. It does not diagnose.
Prime Edge AI
Decides where each AI request is answered: on the device, or by a frontier model in the cloud only when needed.
Gateway Assay
Reviews codebases for technical due diligence, alongside our engineers' expert review.
Draft (F27): Which models and providers we use; any provider that processes data must also appear on the subprocessors page.

Models and providers

[TO CONFIRM]

Human oversight

People review the AI output that matters, and consequential decisions about people stay with people.

Draft (F27): Where people review AI output, and how issues are escalated.

[TO CONFIRM: where review happens, and how issues are escalated]

Telling people when they are dealing with AI

  • Our voice agents say they are AI at the start of every call.
  • Chat assistants label themselves.
  • AI-generated content we publish is labelled where relevant.
Draft (F27): How customer data is used, including whether it is ever used to train or improve models; testing and evaluation (accuracy, safety, quality across languages, fairness); accountability.

How customer data is used

[TO CONFIRM]

Testing and evaluation

[TO CONFIRM: accuracy, safety, quality across languages, fairness checks]

Accountability

[TO CONFIRM: for example the CTO for technical performance, the CSO for clinical safety, the CEO overall]

Known limitations

AI can be wrong. It can misunderstand a request or give an answer that sounds right and is not. That is why people oversee the output that matters, and why our products point people to a person when that is the right next step.

Regulation

  • The EU AI Act's transparency duties for AI systems that interact with people (Article 50) have applied since 2 August 2026.
  • Uses such as risk assessment and pricing in life and health insurance are high-risk under Annex III, with obligations currently due from 2 December 2027.
  • UK GDPR sets rules on automated decision-making, as updated by the Data (Use and Access) Act 2025.
  • The ICO publishes guidance on AI and data protection.

Ask a question or raise a concern

If you have a question or a concern about how we use AI, tell us and we will reply.

Ask about our AI