Full review
Abridge 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 used by clinicians/health systems: records the patient-clinician conversation and drafts the clinical note
Astra review conflicted: OpenAI model in role record.
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 | No | deployer-dependent Not independently replicated | Your clinician will always ask for your consent first.Source This is UChicago Medicine's own policy for its deployment. No Abridge document says a consent or AI notice is always given. Abridge's privacy policy sends patients to their provider's Notice of Privacy Practices. |
| 2. Operator and model maker named | No | not-documented Not independently replicated | leveraging the strongest capabilities of foundation models, like GPT-5.5, to power other featuresSource Abridge AI Inc. is named, and the health system is named in deployments such as UChicago. The model maker is not clearly named for the scribe. This April 2026 blog says Abridge tested OpenAI's GPT-5.5 for note generation and decision support, and mixes in its own proprietary-data models. It does not say which model produces patients' notes. Closest call in this set. |
| 3. Patient chooses to use it | No | documented Not independently replicated | process your data solely on their instructionSource The privacy policy says Abridge is a Business Associate to the provider and acts on the provider's instruction. UChicago's article says the health system rolled it out for clinicians. The patient does not start it. |
| 4. Patient sets the task | No | documented Not independently replicated | record the conversation between you and your clinician and transform it into a clear, accurate clinical note for your EHRSource The purpose (documentation for the clinician and health system) is fixed by the operator. The patient does not set it. |
| 5. Patient can decline or stop | No | deployer-dependent Not independently replicated | If you decline, your visit will proceed just as before, with no change in your care.Source This is UChicago's policy. No Abridge document guarantees a right to decline in every deployment. The Abridge patient terms say use is voluntary, but they cover the retired consumer app, not the clinician scribe. |
| 6. Patient sees the AI output | No | not-documented Not independently replicated | Our software allows clinicians to validate each part of a generated note against the underlying patient conversation transcript and audio recording.Source Sources describe the clinician seeing the AI draft, transcript and audio. UChicago says patients can read the final clinician-approved note through open notes, which is a person-reviewed product. No source says patients see the AI's own draft or transcript. |
| 7. Patient can correct or challenge | No | deployer-dependent Not independently replicated | the right to access, correct, amend or delete such information, please inquire with the relevant Abridge customer directlySource The route runs through the health system, and the policy wording is conditional ("as applicable"). UChicago says patients can message the clinician to fix errors in the note. That is the institution's route, not a vendor route, and it applies to the final note, not the AI draft. |
| 8. Patient can get a copy | No | deployer-dependent Not independently replicated | the right to access, correct, amend or delete such information, please inquire with the relevant Abridge customer directlySource Access is only via the health system. The patient page offered recording export for the consumer app, but that app closed Aug 18, 2025, and the export request deadline of Dec 31, 2025 has passed (https://www.abridge.com/patients). That page does not apply to the clinician scribe. |
| 9. Patient can delete | No | deployer-dependent Not independently replicated | Audio recordings made by the tool are used solely to produce the clinical notes before being deletedSource This is UChicago's retention practice. The patient does not trigger it. Abridge's own policy sends deletion requests to the health system. The consumer-app deletion route (support email) closed Dec 31, 2025. |
| 10. Training or sale does not override patient choice | No | deployer-dependent Not independently replicated | governed exclusively by the terms of our agreements with those customersSource Patient data use is set by private customer contracts that are not public, and no patient-level opt-out is documented. UChicago says "no secondary data use," but that is a deployer statement. The Abridge blog says it trains models on proprietary data, without saying what data. The old consumer terms gave Abridge rights over de-identified derivatives, but they apply to the retired app. |
| 11. Patient can run it or take it elsewhere | No | not-documented Not independently replicated | Abridge processes Customer Data, which may include Protected Health Information (PHI), solely on behalf of our customers.Source Clinician-facing and vendor-hosted. The patient is not a user. The policy points patients to their provider for access, correction and deletion. No self-hosting, no open source, no patient path to take the function elsewhere. |
| 12. No advertising or data business | No | deployer-dependent Not independently replicated | Abridge processes Customer Data, which may include Protected Health Information (PHI), solely on behalf of our customers.Source The policy says use of Customer Data is governed exclusively by Abridge's agreements with each health system. Those agreements are not public, so nothing says the same terms apply to every deployment. Separately, the policy documents that Abridge shared identifiers and device activity with advertising partners for cross-context behavioral advertising. That concerns website and account users, not patient visit data. |
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
Abridge: CAIHL reassessment
September 8, 2026. Published AI-assisted draft.
Abridge is clinician-controlled documentation and decision support. UChicago Medicine documents patient refusal and note-correction routes, but these local commitments cannot describe every customer. Platform expansion does not establish that all patient recordings feed every offering.
Scope and agency posture
Enterprise documentation and clinical decision support. UChicago Medicine is one documented deployment example, not evidence of uniform practice across all customers.
Posture: Mixed.
Axis: 43/100, retained provisionally. No numerical recalibration was performed.
Sources and documented findings
- UChicago states clinicians obtain consent, declining does not change care, patients can review notes and request corrections by message, and audio is deleted within one week. These are local published commitments. https://www.uchicagomedicine.org/forefront/patient-care/ai-ambient-clinical-documentation-what-to-know
- Abridge's CDS page gives clinicians control over information, notes and actions. It promises links to EHR facts and evidence sources and says PHI stays within its architecture and is not used for training. Those CDS-specific claims do not establish every customer's downstream data use. https://www.abridge.com/cds
Patient authority
Inference from the documented workflow: Clinical users decide what enters the record and what action follows. UChicago documents a real patient refusal and correction route, with no claim that the patient controls the assistant itself.
Critical capacity
Editorial assessment: Open notes can support patient checking. Clinician-facing citations do not establish patient access to the same reasoning, recordings or source interface.
Informed control
Editorial assessment: Consent and retention must be described by deployment. Vendor expansion into payer and life-sciences workflows does not prove that the same patient audio feeds every offering.
Assessment and limits
Mixed remains defensible for indirect benefits and locally documented choice. The baseline overgeneralizes platform expansion into a universal audio-use claim. Historical lawsuit allegations should retain dates and must not be presented as current adjudicated facts.
Confidence: Moderate for public deployment policy and CDS design.
Remaining uncertainty: Actual consent practice and downstream use across customers, and current litigation status, were not established
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: 2026-06-12. Historical assessment. Earlier claims are not automatically reverified by this publication.