Responsible AI Policy

Our commitment to responsible AI

At The Heather Cutler Foundation, we believe artificial intelligence can help make pancreatic cancer information, support and services easier to access.

But technology must never come at the expense of patient privacy, dignity, safety or trust.

We use AI to help people find information and navigate available support. We maintain human oversight of consequential decisions, protect personal information, prioritize reliable sources and remain accountable for the services we provide.

Our commitment is simple: technology should make it easier to access care and support, never harder to trust it.

What this policy covers

This policy describes how the foundation uses artificial intelligence in its public services, what those systems are not permitted to do, and how people oversee them. It is written to align with the six principles in the September 2026 joint resolution of Canada's federal, provincial and territorial privacy regulators on the responsible use of AI in health and health research: privacy protective, valid and reliable, safe, human rights affirming, transparent, and accountable. That resolution is regulator guidance, not a law, and this policy is a statement of our practices — not a claim of legal compliance. The foundation's handling of personal information is governed by applicable Canadian law, including PIPEDA and, in Newfoundland and Labrador, the Personal Health Information Act (PHIA), and by our Privacy Policy.

How we use AI today

We describe our systems as they actually are, including their limits.

Canadian clinical trials search (live, in testing). Our trials service at heathercutler.ca/mcp/trials lets a person's own AI assistant search Canadian pancreatic cancer trial listings. The server itself contains no language model: it answers database queries against a mirror of ClinicalTrials.gov listings, refreshed daily, and returns records with their source links and dates. It cannot invent a trial, a site, or a date. No search parameter accepts personal health information — there is nowhere to enter it, so none is collected or stored. Responses carry a TEST BUILD notice while the service is in testing.

Structuring trial criteria (behind the scenes). Eligibility criteria on ClinicalTrials.gov are free text. Automated tools — including, since November 2026, a language model working inside the foundation's own systems on public registry text — help our team structure those criteria so they can be searched. Everything those tools produce is an unverified proposal: nothing derived from criteria is used to filter or present results until a person has checked it against the published criterion, word for word. No model is involved when a person or an assistant queries the service.

Port to Care applications (live pilot). Port to Care is the foundation's travel-grant program and its only live patient program. Its application service at heathercutler.ca/mcp/port-to-care also contains no language model: it validates a small set of fields, shows the applicant a plain-language summary of exactly what will be submitted, and submits only after the applicant confirms the summary and records consent. It collects the minimum needed to process an application — contact details, community and province, travel purpose and approximate dates, and the amount requested (grants are up to $500 CAD). It does not ask for or accept diagnosis details, health card numbers, income information, or medical records. Confirmation emails contain no medical detail. Every application is reviewed, and every grant decision is made, by a person at the foundation. The email route to apply remains open alongside this service.

Site search (interim). A search service over our published pages (Cloudflare AI Search) returns page text to assistants. It is a plain text pipe with no verification labels; we describe it as interim. Its replacement — a governed knowledge service whose records carry sources, checked dates, and verification status — is in design and will be launched in stages as content is verified.

Trial matching (in design). We are designing a way for a person's own assistant to compare trial criteria against information the person gives that assistant. The design rule is that the profile never reaches us: our server receives anonymous filter parameters only (for example, a mutation name and a province), never a person's details or health information. The strongest statement the service will ever produce is "you may qualify for this trial" — the trial team makes the final determination at screening.

Internal use. Foundation staff and volunteers use AI tools to assist with drafting, research, and administration. Nothing produced with AI assistance is published, sent, or acted on in the foundation's name without human review and approval.

What our AI systems never do

• Decide or predict whether a person is eligible for a clinical trial, or whether a grant application will be approved. People make those decisions.
• Give medical advice: no treatment recommendations, dosing, comparisons of treatments, or prognosis.
• State that a drug is approved, funded, or available in Canada when the data does not say so.
• Accept, store, or repeat personal health information in our trials and knowledge services — they are built with no field that can hold it.
• Present content as medically reviewed when it is not. Our guides carry "Pending medical review" until a qualified medical professional has actually reviewed them, and that status travels with the content wherever our services carry it.
• Present a registry listing as verified when it is not. Records a person has checked are labelled verified; everything else is labelled a registry mirror.
• Invent or pad results. A search with no matching records returns an honest empty result and the route to a human.
• Claim a partnership with another organization that the foundation has not confirmed.

Human oversight

• A named person is accountable for every service: the Executive Director owns the governance documents, and changes to them require his approval.
• Verification is human. Trial listings and structured criteria carry a verified label only after a person has checked them against the source. Model output never verifies anything, including its own.
• Every service has a kill switch, and disabling a service takes precedence over keeping it available. A suspected privacy incident stops the service first; it is reported to the Executive Director the same day.
• Before changes ship, services must pass fixed test sets — including requests the service exists to refuse, such as eligibility decisions and medical advice — and produce the constrained response or the route to a human, never an improvised answer.
• Incidents are recorded in the governing document, with the date and the cause.

Sources and citation

Our published research is meant to be attributed, not detached from its source. Where You Live Matters: How Canada Responds to Pancreatic Cancer (2026) is licensed CC BY 4.0 — share and adapt freely, with credit to The Heather Cutler Foundation. Records our knowledge services return carry the publisher, the work's title, the section, the publication date, and a direct link to the source page, so that any assistant repeating our findings can say where they came from. Trial records carry the ClinicalTrials.gov source link and both the registry's last-updated date and our last-synced date, so staleness is visible in the answer itself.

Raising a concern

If you believe one of our AI services has given wrong information, handled information improperly, or behaved in a way this policy does not permit, contact us at info@heathercutler.ca or 1-855-754-5045. Concerns are read by a person. You can also reach the foundation about any application or service at support@heathercutler.ca. Withdrawing consent for a Port to Care application is done by emailing support@heathercutler.ca; withdrawing consent may end the application.

Review

This policy is reviewed at least annually, and whenever we launch a new AI system or change an existing one materially. Each review updates the version and date above. The Support Line and the ABAP biomarker-testing program are in development; when any new patient-facing service launches, this policy and the foundation's governance documents are updated first, and the service does not launch on the strength of this draft.