Patient-First AI: Felix’s principles for AI in your care

At Felix, patients come first, and that includes how we use AI. These seven principles govern every way Felix uses AI in patient care. Ask Felix, our AI assistant that pairs with your Felix care team, is the first product built on them. The principles set out how we build our AI, what it does and does not do, and who answers for it when something goes wrong.

Each principle has two parts: what we believe, and what this means in Ask Felix.

1. First, do no harm

What we believe. Patient safety comes before every other goal, including speed, convenience and engagement. We take this principle from medicine because the stakes are the same.

Ask Felix is patient, kind and non-judgmental. It tells patients the truth, not what they want to hear. When it is uncertain, it says so. When a question needs a human, it hands it off instead of guessing.

Ask Felix has been built to operate within applicable law, Felix policies and our terms of service. 

What this means in Ask Felix.

• Every message goes through 17 safety checks before it reaches the main AI model. Five of those checks can end the conversation before the main model is ever used.
• Layered safeguards check every response for unsafe, inaccurate or inappropriate content before it reaches a patient, and we continue to review responses after they are sent.
• When a patient describes symptoms that may be urgent, Ask Felix stops the conversation and directs them to emergency services or a practitioner. It does not triage or indicate that seeing a practitioner is not necessary. 
• Ask Felix says “I’m not sure” or “This is a question for your practitioner” rather than present a guess as fact.
• If Ask Felix makes a claim about a patient’s health or treatment, it substantiates that claim against the patient’s actual Felix health record, with their permission.
• Clinical and security experts test Ask Felix before launch and before every significant change, including deliberately trying to get around its safeguards.

2. Support human care, never replace it

What we believe. Ask Felix exists to make care from real practitioners more accessible and patient-centred. It helps patients talk through symptoms, understand treatment options, navigate their care and prepare for appointments. It does not diagnose, prescribe or make clinical decisions. Those activities belong to qualified practitioners.

This is what sets Ask Felix apart from a general-purpose chatbot. It pairs with your Felix care team, so it is a pathway to real care, never a dead end.

What this means in Ask Felix.

• Ask Felix can hand a patient off to a Felix practitioner at any point in the conversation, and the conversation’s context goes with them. The patient never has to start over.
• Ask Felix does not diagnose a condition, write or change a prescription, or tell a patient to start or stop a medication.
• Patients can always message their practitioner directly if they prefer. Ask Felix is never a required step to reach a human.
• Ask Felix is designed with clear boundaries around what it can and should do, and it enforces those boundaries by deferring to a practitioner for assessment.

3. Personal to you, controlled by you

What we believe. The most helpful AI understands the person it is helping. When a patient chooses to share their information, Ask Felix can tailor its answers using their medical history, medications, treatments and past conversations with their care team.

If a patient does not want to share that information, Ask Felix cannot access it but can still provide general assistance. The patient remains in control of what information Ask Felix can use.  

What this means in Ask Felix.

• Patients decide whether Ask Felix uses their health information, and they can change that decision at any time.
• Patient data is never used to train AI models. No exceptions. Our AI providers operate under zero-data-retention terms that prevent patient data from being retained for model training.
• Ask Felix uses personal characteristics only when they are relevant to the patient’s care or question, rather than using them to make assumptions about the patient.
• When Ask Felix helps a patient take an action, such as updating their information or managing their account, it does only what the patient is already allowed to do, with the same safeguards as if they did it themselves.
• Ask Felix confirms with the patient before it takes any action on their behalf.

4. Always clear that it’s AI

What we believe. Patients always know when they are talking to AI, what it can and cannot do, and when a human practitioner is involved (or when they should be). Ask Felix is honest about the limits of what it knows and never sounds more confident than the evidence allows.

Transparency covers data too. Patients understand what information Ask Felix sees, how it uses that information to personalize their experience, and what choices they have.

What this means in Ask Felix.

• Ask Felix is clearly labelled as AI. It never presents itself as a doctor, nurse practitioner, pharmacist or any other member of the care team. It is not intended to replace the clinical judgement of a health care practitioner.
• When a conversation needs a practitioner, Ask Felix helps connect the patient to their care team and makes it clear that they are now speaking to a human.
• Patients are informed about what information Ask Felix has access to, how it may be used and what choices they have about sharing it.
• We publish a plain-language explanation of how Ask Felix works, what it is designed to do and where its limits are.
• Ask Felix is never a required step to reach a healthcare practitioner. 

5. Always measuring, always improving

What we believe. AI changes quickly, and Ask Felix improves with it, safely. We systematically measure how safe, accurate and helpful Ask Felix’s real-world responses are, and we find where patients are not getting the help they need.

As Ask Felix gains new abilities and knowledge, our safeguards and testing grow with it. A capability is not ready until we can test, monitor and control it.

What this means in Ask Felix.

• Every Ask Felix conversation is anonymized and automatically reviewed against defined quality and safety criteria. Conversations that don’t meet those criteria are escalated for human review.
• Every new capability is tested before it reaches patients, and we re-run our safety and quality tests when the underlying AI model changes. 
• Patients can flag any response that is wrong or unhelpful, and every report feeds directly into our review process.

6. Felix is accountable

What we believe. Felix is responsible for Ask Felix and every experience it creates. We decide what it is allowed to do, put safeguards around those abilities, evaluate how it performs, watch how it behaves in the real world, and step in when something is not working.

We acknowledge that AI can make mistakes. Felix remains accountable for Ask Felix, including how it performs, how issues are addressed and how we work to reduce the risk of those issues recurring.

What this means in Ask Felix.

• Named people at Felix own Ask Felix’s safety, clinical quality and privacy, and they have the authority to change or pause it.
• We can quickly switch off any feature that is not working as intended.
• We review these principles at least once a year, and whenever Ask Felix gains a significant new capability or when laws and industry standards change.

7. Private and secure by design

What we believe. Health information is some of the most sensitive information a person has. Ask Felix meets the same privacy and security standards as every other part of the Felix platform.

Ask Felix complies with the Personal Information Protection and Electronic Documents Act (PIPEDA) and with the provincial private-sector and health information privacy laws that apply where Felix operates. 

Privacy and security commitments. Ask Felix collects, uses and discloses personal information only for the purposes patients consent to, and only to the extent necessary. Physical, organizational, contractual and technological safeguards protect it against loss, theft and unauthorized access, use or disclosure. Only people and service providers with a genuine need to know have access, and all of them complete privacy and security training and are bound by our policies. Information is stored in Canada. The same privacy rights that apply to other information held by Felix also apply to information used by Ask Felix. No system is risk-free, so we review our safeguards continuously as Ask Felix evolves.

What this means in Ask Felix.

• Ask Felix sees only the information a patient chooses to share with it, and uses it only to help that patient.
• Conversations with Ask Felix carry the same security controls as the rest of a patient’s Felix account.
• Privacy questions and requests about Ask Felix go to the Felix Privacy Officer, like any other privacy request.
• We conduct regular adversarial security testing, including testing for prompt injection and other AI-specific attack techniques, and address issues identified through that testing.

How we govern AI at Felix

An AI governance committee, with representatives from clinical, compliance, security, and product, approves every new AI capability before launch and reviews its performance after.

• Incidents. We triage, investigate and fix every safety, bias or privacy incident, and we document the root cause. We notify affected patients and regulators wherever the law requires it.
• Third-party models. We assess every AI provider for safety, privacy and security before use, and we hold them to the same standards as Felix.
• Reporting. Felix is committed to publishing an annual Patient-First AI report on how Ask Felix performs against these principles.

Please reach out to us at [email protected] with any questions.