How Heard handles privacy, consent, AI, and care.
Adapted from the principle frameworks behind the Belmont Report, the Declaration of Helsinki, and Cambridge's research integrity guidance — restated in language that fits a consumer healthcare web app rather than a research lab.
Version 1 · last reviewed June 2026. This page changes when our practice changes.
Core principles
- 1Respect for persons
Users are autonomous decision-makers about their own bodies and their own data. Heard's role is to surface evidence — never to make decisions for someone, replace clinical judgement, or override consent.
- 2Beneficence — do good, minimise harm
Every feature is designed to reduce diagnostic friction without amplifying anxiety. We deliberately avoid alarmist framing, do not score users on health, and surface 'worth asking about' rather than 'you have'.
- 3Justice — fair access
Heard is free, account-less, and works on any device with a browser. We do not charge for advocacy. We do not paywall language, accessibility, or core features.
- 4Integrity & transparency
We disclose how every output is generated (local heuristics, browser LLM, or Claude API). We cite sources for statistics. We do not invent symptoms, fabricate evidence, or produce content the user did not log.
- 5Accountability
Source is open. Bugs and ethical concerns can be raised on the project repository. Material design decisions (AI defaults, language choices, condition lists) are documented in code and can be audited.
UN Sustainable Development Goals
Of the 17 UN SDGs, Heard works directly on three. The choice is not incidental — these three are the goals our ethical principles translate into measurable outcomes.
Diagnostic delay is a public-health problem. Better-prepared appointments produce faster, more accurate diagnoses. Heard's brief gives a clinician 90 seconds of dense, time-stamped evidence in place of a vague oral history.
Women wait 7+ years for endometriosis. They are 50% more likely to be misdiagnosed after a heart attack. 80% of autoimmune disease occurs in women. Heard names the gendered structural cause of these gaps and gives users the data infrastructure to push back against it.
Heard is free, account-less, and runs in any browser. No language paywall, no premium tier for advocacy, no requirement to share data with a third party. Self-advocacy infrastructure should not cost money.
The other 14 SDGs are visible on the home page for context. Heard does not claim to work on goals it does not work on.
Specific commitments
Data & privacy
- Symptom entries live in the user's browser localStorage by default. There is no account, no server-side database, no third-party analytics, no advertising pixels.
- The optional Claude API mode sends a compacted symptom log to Anthropic only at the moment a brief is generated, and only on explicit user action. No identifying fields (name, email, location) are sent.
- The browser-side local model (Llama 3.2 1B via WebLLM) runs entirely on-device. Symptom data never leaves the browser when this mode is selected.
- We do not sell, share, or repurpose user data. There is no data to sell.
Informed consent
- Every AI mode is opt-in. The default brief generator is fully local heuristics — no AI, no network call.
- The dropdown on /brief clearly states what each mode does and what data crosses the network. Users choose with eyes open.
- Demo data is clearly labelled and can be wiped from the dashboard at any time.
AI use & limitations
- Outputs are framed as 'worth asking your clinician to consider', never as diagnoses.
- Where AI augments a brief, the source is named on the output (Claude Sonnet 4.6 / Local model / Local heuristics).
- We do not use AI to make decisions about treatment, eligibility, risk-scoring, or insurance. Heard is an advocacy tool, not a triage system.
Inclusivity & language
- Heard is written for women, non-binary people, and people assigned female at birth (AFAB). The term AFAB is spelled out on first use throughout the app.
- We avoid trans-exclusionary language and we name the diagnostic gap as a gendered structural issue, not a personal failing.
- The condition library and pattern engine are reviewed for the dismissive idioms that women's medicine has historically used ("anxiety", "sensitive", "normal for women").
Vulnerable users & trauma-aware design
- Many users come to a symptom journal after being dismissed by a clinician. The "previously dismissed" flag is a first-class field, not a footnote, and it informs the brief.
- We do not surface emergency-room or self-harm content without context. If a symptom pattern is potentially urgent (atypical cardiac signals, fainting), Heard recommends contacting a clinician, not a triage chatbot.
- There is no streak counter, no engagement metric, and no gamification. The app should not reward suffering with points.
Safety & dual-use
- Heard is an educational self-advocacy tool. It is not a medical device, it is not CE-marked, FDA-approved, MHRA-registered, or regulated as software-as-a-medical-device.
- We will not add features that diagnose, treat, prescribe, or claim clinical efficacy.
- We will not integrate with insurance underwriters, employer health programmes, or other systems that could weaponise symptom data against the people who logged it.
Conflicts of interest
- Heard has no commercial relationship with pharmaceutical companies, insurance providers, or device manufacturers.
- The Anthropic dependency is disclosed; users can run the app without it.
- Sponsorship or grant funding, when accepted, will be disclosed on this page.
Accessibility
- Heard is mobile-first, responsive, and uses semantic HTML with ARIA labelling on interactive elements.
- Body copy is set at a comfortable reading size with sufficient colour contrast against the warm background.
- We will treat accessibility regressions as bugs, not as nice-to-haves.
Raising concerns
- If something in Heard feels unsafe, inaccurate, or exclusionary, please open an issue on the GitHub repository or contact the team via the SupportHers.com channels.
- We treat ethics issues as P0 — equal priority to security incidents.
What Heard is not
Heard is not a medical device. It does not diagnose, treat, cure, or prevent any condition. It does not replace consultation with a qualified clinician. It does not provide emergency advice — if you are in a medical emergency, contact your local emergency services.