Full review
Ada Health CAIHL draft report
Evidence-linked HugoScore draft report for a health AI tool that affects patients.
Agency checklist · October 30, 2026
Leans expands 8 of 12 yes
Document-based draft. Independent confirmation is incomplete. Hugo has not personally verified this service.
Assessed function: Direct Ada symptom assessment (Ada Assess) used by the patient on ada.com or the Ada app. Partner-embedded Assess and the US advertising-segment consent are distinguished in notes.
A no can mean missing documentation or institution-dependent control; it does not by itself establish harmful behavior. These categories are not clinical safety or product-quality ratings.
- Public documents do not establish actual product behavior.
- The independent review screened rules and selected evidence; it did not replicate all twelve answers for this tool.
See all twelve answers and their evidence
| Question | Answer | Evidence / review | Source and reasoning |
|---|---|---|---|
| 1. Told it is AI | Yes | documented Not independently replicated | Ada Assess is an AI software application that uses custom probabilistic reasoning technology on a medical-expert-curated knowledge baseSource Terms section 1.3 also describes 'an in-house developed, artificial intelligence (AI) model'. Last modified 6 Nov 2025. |
| 2. Operator and model maker named | Yes | documented Not independently replicated | Ada utilizes the services of Together AI, a third party LLMSource Ada Health, Inc. (US) and Ada Health GmbH are named. The core model is Ada's own PGM. Together AI (LLM) and Microsoft Azure (free-text input) are named. |
| 3. Patient chooses to use it | Yes | documented Not independently replicated | Ada Assess is available for free via our website and (when and where available) via websites/apps of business partners.Source Direct use is the user's choice. In partner-embedded versions the partner chooses to deploy it, and I did not assess that variant. |
| 4. Patient sets the task | Yes | documented Not independently replicated | to provide Users with health information relevant to medical findings they are currently experiencingSource The user supplies the symptoms. The fixed function is symptom assessment, with Ada's questions and an optional partner-care step. |
| 5. Patient can decline or stop | Yes | documented Not independently replicated | You are entitled to stop using this website, Ada Assess, Ada PHC and your User Account at any timeSource Self-started tool per the Q3 source, no care depends on it. |
| 6. Patient sees the AI output | Yes | documented Not independently replicated | the User is provided with a report that lists such information as the conditions likely to be causing these findingsSource The Ada Health GmbH policy says the report can be downloaded to the device. |
| 7. Patient can correct or challenge | No | not-documented Not independently replicated | Quotation not recorded; evidence repair pending. Source The policies list GDPR/CCPA rights to rectify personal data, but no route to correct or contest the assessment output is documented. |
| 8. Patient can get a copy | Yes | documented Not independently replicated | You are not required to receive the report this way and can instead download the report onto your device.Source This is the Ada Health GmbH (app) policy. It also lists data portability rights. |
| 9. Patient can delete | Yes | documented Not independently replicated | you may request the deletion of a specific case or delete the case yourselfSource The GmbH policy says data is deleted or irreversibly anonymized within 1 month. The US terms say deactivation deletes personal data but Ada keeps rights to anonymized data. |
| 10. Training or sale does not override patient choice | No | documented Not independently replicated | To improve our Services, as well as for training and quality purposesSource The US policy says third-party LLMs will not train on user data. But terms 5.2 say Ada uses Together AI 'for internal AI training', the policy lists developing 'Artificial Intelligence/Large Language Models', and the GmbH policy trains on anonymized assessment data. No opt-out is documented. |
| 11. Patient can run it or take it elsewhere | No | not-documented Not independently replicated | Quotation not recorded; evidence repair pending. Source Vendor-hosted only. |
| 12. No advertising or data business | No | documented Not independently replicated | Ada also creates advertising segments which it sells to its marketing partners.Source US policy section 6.2 makes this consent-based, but the rule counts it even with consent. Terms 3.2 grant Ada a perpetual license to use anonymized data 'in any manner'. Conflict: the GmbH app policy says Ada will 'never share your personal health information with advertisers'. The US policy governs for US users. |
Initial coding: Claude Sonnet helpers, compiled by Claude Opus 5.5 (Anthropic). Selective review and corrections: GPT-6 Astra (OpenAI). Editor: Hugo Campos. Acceptance of corrections is not product verification.
Checklist rules · Independent review and limitations · Later evidence follow-up
Earlier report
Ada Health: CAIHL reassessment
September 8, 2026. Published AI-assisted draft.
Ada’s direct assessments differ from partner-controlled sharing and decisions. Its US policy describes consented health-derived marketing segments. Removing an existing segment requires a separate request, beyond changing the consent setting.
Scope and agency posture
Direct Ada assessment, partner-embedded Assess, and US advertising-segment consent must be distinguished.
Posture: Mixed, potentially agency-expanding.
Axis: 66/100, retained provisionally. No numerical recalibration was performed.
Sources and documented findings
- The current policy says patients may share a report with clinicians. Partner deployments may require assessment sharing and may make their own automated decisions, with rights directed to the partner. https://ada.com/privacy-policy/
- The US policy describes consented marketing segments derived internally from assessment outcomes and represented by pseudonymized tokens. Revoking the consent setting alone is insufficient to remove an existing segment: an email-address-based request is required. https://ada.com/us/consumer-health-data-privacy-policy/
Patient authority
Inference from the documented workflow: Direct assessment can support self-directed care seeking. Mandatory sharing in some integrations is an actual deployment condition, while partner routing decisions are not attributable to every consumer assessment.
Critical capacity
Editorial assessment: Reports support reflection and action. Correcting personal data and contesting an institution's consequential decision are separate routes, and actual report-level revision remains untested.
Informed control
Editorial assessment: The separate segment-removal action materially affects meaningful withdrawal. Pseudonymization should not be described as the absence of health-derived commercial targeting.
Assessment and limits
Mixed remains warranted from documented conditions. Update the rationale with the specific withdrawal burden rather than a generic pharma-funding concern.
Confidence: Moderate for policy distinctions.
Remaining uncertainty: In-app consent comprehension, report correction and partner alternatives remain unverified
Published documentation is evidence of stated conditions, not proof of actual implementation. Unknowns did not receive automatic negative points. Funding, sponsorship, and public code do not determine agency by themselves.
Review provenance
- Reviewer/model: OpenAI Codex / GPT-6.
- Method: Focused public-source reassessment using CAIHL: patient authority, critical capacity, and informed control. Existing evidence plus one focused primary-source pass and at most one targeted follow-up. No live product testing. Numerical scores remain provisional editorial placements, not a new calculation.
- Human review: Hugo Campos authorized publication of these AI-assisted draft reassessments on September 8, 2026. This does not claim comprehensive human verification of every finding.
- Earlier review: 2026-06-30. Historical assessment. Earlier claims are not automatically reverified by this publication.