HugoScore hugoscore.org

Full review

Nabla CAIHL draft report

Evidence-linked HugoScore draft report for a health AI tool that affects patients.

Agency checklist · October 30, 2026

Constrains 0 of 12 yes

Document-based draft. Independent confirmation is incomplete. Hugo has not personally verified this service.

Assessed function: Ambient AI scribe for clinicians: records the patient-clinician conversation and drafts the clinical note

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
QuestionAnswerEvidence / reviewSource and reasoning
1. Told it is AINodeployer-dependent
Not independently replicated
Some state and federal laws require patient consent to use ambient scribing solutions. Source

Nabla publishes a sample consent script titled "Patient consent example." Each clinic decides whether to use it. No Nabla document says the patient is always told.

2. Operator and model maker namedNonot-documented
Not independently replicated
internal artificial intelligence model Source

Nabla is named. The terms say Nabla trains its own "internal" model, but they do not say what model produces the notes. The Trust Center lists Google Cloud (hosting) and Azure (speech-to-text) as subprocessors and no language-model vendor. The Nabla AI Model Card is available only on request, so I could not read it.

3. Patient chooses to use itNodocumented
Not independently replicated
designed and intended to be used by healthcare professionals and other authorized users Source

The service is contracted by and for clinicians. The patient does not start it. The consent script has the clinician tell the patient they are using the tool.

4. Patient sets the taskNodocumented
Not independently replicated
This tool will listen to our conversation and convert it into a summary Source

The purpose (a clinical note) is fixed by the clinician and operator.

5. Patient can decline or stopNodeployer-dependent
Not independently replicated
Do you consent to the use of this tool during our visit and future visits? Source

This is an example script a clinic may or may not adopt. Nabla's documents do not state that a patient who declines still gets the same care.

6. Patient sees the AI outputNonot-documented
Not independently replicated
convert it into a summary, which I will review and edit for accuracy Source

The clinician reviews and edits the summary. The consent script says the audio is not saved. No source says the patient sees the AI's draft or transcript.

7. Patient can correct or challengeNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

No patient route to correct or contest the AI output is documented. Review is by the clinician only.

8. Patient can get a copyNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

The privacy policy lists access and portability rights. It defines its scope as website visitors, authorized users and email recipients, which does not clearly include patients. The Trust Center says audio is not stored by default.

9. Patient can deleteNonot-documented
Not independently replicated
We retain clinical notes for a short period of time (14 days), which is configurable by client Source

Retention is automatic and set by the client. The patient has no documented way to trigger deletion. The privacy policy's erasure right has the same scope ambiguity as Q8.

10. Training or sale does not override patient choiceNodocumented
Not independently replicated
Nabla may collect, process, and otherwise use de-identified data that is transmitted to Nabla through your use of the Services for training Source

Conflict: the binding terms let Nabla train on de-identified data from service use, with no patient opt-out. The terms add that "audio data does not include patient audio." The privacy policy also allows sharing de-identified health information with third parties. A separate Nabla page (https://nabla.com/copilot-audio-patient-consent) says patient consent is required to share audio, note and transcript to improve the service. That page is a sample script, and the terms take precedence per the rules.

11. Patient can run it or take it elsewhereNonot-documented
Not independently replicated
No portion of the Services may be reproduced in any form or by any means. Source

Clinician-facing SaaS under a user license. No self-hosting or patient portability path in the terms, privacy policy or trust page.

12. No advertising or data businessNodocumented
Not independently replicated
Nabla may disclose such de-identified data in connection with the Services. Source

The same clause says Nabla may collect and use de-identified data for training its internal AI model. The attached BAA lets Nabla de-identify PHI for product improvement. The clause states that audio data does not include patient audio.

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

Nabla: CAIHL reassessment

September 8, 2026. Published AI-assisted draft.

Nabla gives clinicians and healthcare organizations control over documentation, coding, and retention settings. Patient access, refusal, and correction depend on the deployment. Earlier consent and training assurances were not fully reverified in this pass.

Scope and agency posture

Nabla's clinician-controlled ambient documentation, dictation, and coding workflows within provider organizations. Patient-impacting institutional AI.

Posture: Mixed.

Axis: 40/100, retained provisionally. No numerical recalibration was performed.

Sources and documented findings

  • The current product homepage places note customization and final coding decisions with clinicians and advertises organization-configurable retention.
  • The public Trust Center says audio is not stored by default, notes are retained 14 days subject to client configuration, and optional clinician feedback may contain de-identified data. Its model card and audit documents require requests and were not accessed.
  • The prior product and patient-consent URLs returned non-retryable access errors. A single homepage follow-up confirmed the current function, but not the old consent wording.

Patient authority

Inference from the documented workflow: Clinicians and organizations choose the workflow and retention settings. A patient may benefit without controlling the note's purpose, final content, or coding. The evidence supports this allocation, not a judgment from vendor ownership.

Critical capacity

Editorial assessment: Reviewed documentation may improve later understanding, but no direct patient mechanism to inspect or contest generated notes was established. Coding traceability for clinicians is not automatically patient-accessible.

Informed control

Editorial assessment: Concrete retention and hosting disclosures are useful. The baseline's no-training-on-notes and consent claims were not fully reverified and should remain dated claims rather than universal guarantees.

Assessment and limits

Mixed remains appropriate. Available governance documentation supports conditional safeguards but does not show transfer of final authority to patients.

Confidence: Medium for institutional deployment and retention statements, low for patient-facing rights.

Remaining uncertainty: Current consent guidance, no-training promise scope, local refusal alternatives, transcript access, and patient correction mechanisms.

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: 2026-06-08. Historical assessment. Earlier claims are not automatically reverified by this publication.