Security Policy

PulseLife evidence-based medical AI, built for safety.

🖇️ Access the resources for safe use of PulseLife's Evidence-based Medical AI: 

Ask our Medical AI Expert your questions

Safety is a critical issue in healthcare. 

LLMs are probabilistic models. They're not designed to give the certainly correct answer, but the most probable one for each question - which is why we talk about AI "hallucinations." 

The challenge for AI technologies is to work towards "zero hallucination." 

PulseLife's AI Expert was built with this goal at its core, to guarantee complete safety. 

Our AI safety policy rests on six fundamental, complementary pillars that guide every technological and functional choice behind the solution: locally grounded clinical sources, secure hosting, reliability, transparency, human oversight, and regulatory compliance. 

📁 A living, trusted and recognised knowledge base 

PulseLife's AI Expert draws on a proprietary, multi-source knowledge base and prioritises the sources that carry the most authority in each region (currently set at country level). 

Globally, this base has been built up over more than 10 years and now holds over 5,000,000 pieces of scientifically recognised, trusted content, with sources validated by our Scientific Council. Since launching in the UK, we've built a dedicated UK source base from the ground up with hundreds of clinical sources already curated by our clinical team, and growing every week. 

Content is updated automatically every day, with manual human review. 

1

Controlled, adaptive technology, built to get the best out of AI 

Unlike almost all so-called "medical AI" tools, PulseLife uses a RAG-LLM: a technology that restricts the AI model to querying PulseLife's knowledge base, and only that base. 

As a result: 

→ Clinicians can be confident, at every stage, that each answer draws on the full body of currently published reference material, and nothing else. 

→ For every paragraph of a response, the model states the source content, its exact publication date, and a link for verification. 

PulseLife's AI is therefore fully traceable in detail and can be supervised by the user at every step. 

2

An LLM layer that adapts at any time 

The LLM layer of our technology, responsible for generating text, is model-agnostic: our system can run on any LLM on the market. Response quality and data protection standards are the key criteria behind each choice of LLM. In a fast-moving field, PulseLife isn't tied to any single provider, and regularly reviews its choice of supplier. 

PulseLife's Medical AI always prioritises a safe answer over a fast one.

🤐 An AI bot should know when to stay silent 

Doubt is a virtue in medicine; it's at the heart of the search for the truth. When a GP is uncertain, they refer on to a specialist. That's what modern medicine is built on. Making sure the AI bot declines to answer when asked about conditions that don't exist, or about management with insufficient scientific evidence, is an essential safety mechanism, and one of PulseLife's key strengths. 

💬 Have a question? Contact us here 

🤖 An LLM layer that adapts at any time 

The LLM layer of our technology, responsible for generating text, is model-agnostic: our system can run on any LLM on the market. In a fast-moving field, PulseLife isn't tied to any single provider, and regularly reviews its choice of supplier. PulseLife currently offers two AI models: 

  • The standard model runs on Gemini: PulseLife currently considers this to offer the strongest clinical evidence base for safety. 
  • The secure-hosting model runs on Mistral: under this model, PulseLife operates under a Zero Data Retention (ZDR) contract. Mistral does not store, use or analyse any data passing through its model, and no data is used to train the models. This contractual guarantee meets the highest industry standard for data protection. 

🔒 Independently certified, safeguarded hosting 

PulseLife's infrastructure is independently certified and rigorously governed. Data is hosted with Scaleway, certified to HDS (the French national standard for health-data hosting), and one of the most demanding in the world. No data is subject to the US Cloud Act, FISA, or any other US legislation. Data is hosted within the EU, under full GDPR-aligned safeguards. 

🏛️ The six pillars of safety 

Pillar 

Description 

What this means in practice 

Locally grounded 

Answers rooted in recognised UK clinical sources 

• UK-recognised sources prioritised first, with broader European and international evidence used as a fallback where local coverage doesn't exist 

Securely hosted 

Certified, safeguarded infrastructure 

• HDS-certified hosting, GDPR-aligned safeguards 
• No exposure to the US Cloud Act, FISA or other US data laws 

Reliable 

RAG-LLM built on a tightly controlled source base 

• A fully controlled source list, published in full 
• An effective update system 
• Fine-grained annotation of every piece of content 

Transparent 

Building trust: say what you do, and do what you say 

• Source date and clickable link on every answer 
• Server location disclosed 
• LLM in use disclosed 

Supervised 

Continuous human oversight of the technology 

• Human validation of source-base content 
• An alert system built on test questions, checking both what's sent to the LLM and the answer it returns 
• Clinicians can apply their own critical judgement, via an evidence panel justifying each answer, with sources clickable at paragraph level 

Compliant 

Meeting regulatory requirements 

• GDPR 
• HDS (the French national standard for health-data hosting) 
• EU AI Act 

❓ A question or comment about PulseLife's safety approach? 
Contact us hereÂ