HugoScore hugoscore.org

Full review

Microsoft Dragon Copilot CAIHL draft report

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

Agency checklist · October 30, 2026

Constrains 2 of 12 yes

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

Assessed function: Microsoft Dragon Copilot clinician documentation and nursing workflows (dictation, ambient note generation, flowsheet capture, generative drafts), deployed by healthcare organizations. Institutional AI affecting the patient record, not a patient-operated assistant.

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
QuestionAnswerEvidence / reviewSource and reasoning
1. Told it is AINodeployer-dependent
Not independently replicated
Requires patient consent (per organization policy) for recording. Source

Microsoft's own terms make the customer solely responsible for notices and consents (Service Specific Terms). No patient-facing AI notice is built in by Microsoft that the sources show. Whether a patient is told depends on the deploying organization.

2. Operator and model maker namedYesdocumented
Not independently replicated
Dragon Copilot is powered by Microsoft’s conversational and ambient AI capabilities Source

Microsoft is named as the operator and AI maker. The Transparency note also names Azure OpenAI Service content filters. The deploying clinic or health system that actually runs it in a given encounter is not named by Microsoft and varies.

3. Patient chooses to use itNodeployer-dependent
Not independently replicated
only intended to be used by a qualified member of an individual's care team Source

Use is by the clinician on the customer's behalf. The patient does not start or choose it.

4. Patient sets the taskNodocumented
Not independently replicated
The clinician controls all decisions related to healthcare Source

The clinician and organization set the task and outputs (notes, referral letters, after-visit summaries). The patient does not set the purpose.

5. Patient can decline or stopNodeployer-dependent
Not independently replicated
if a patient prefers not to have the encounter itself be recorded Source

Microsoft documents that a clinician can dictate after the visit instead of recording the encounter. That is narrower than declining AI-assisted documentation, since dictation still produces an AI draft. Recording consent is per organization policy.

6. Patient sees the AI outputNodocumented
Not independently replicated
The generated clinical documentation is returned to the clinician for review. Source

Output goes to the clinician. The patient sees at most a clinician-reviewed after-visit summary, not the AI's own output.

7. Patient can correct or challengeNodeployer-dependent
Not independently replicated
how Microsoft can assist the customer in fulfilling data subject rights Source

Any route is through the customer (covered entity). No patient-facing correction or contest route is documented by Microsoft.

8. Patient can get a copyNodeployer-dependent
Not independently replicated
The customer is solely responsible for downloading, exporting, and permanently storing all relevant Dragon Copilot data Source

Retention is up to 90 days (FAQ) and the customer files to the EHR. The patient has no direct access to a copy.

9. Patient can deleteNodeployer-dependent
Not independently replicated
a patient revokes previously provided consent, a customer can log a ticket with customer support Source

Deletion of identifiable data after consent revocation is initiated by the customer, not the patient. The same page says anonymized data created before revocation will not be purged.

10. Training or sale does not override patient choiceNodocumented
Not independently replicated
Dragon Copilot's AI/ML models are trained solely on anonymized data. Source

Customer Data is anonymized within 90 days and used to improve models. No patient opt-out is documented. Opt-out exists only as the customer purging identifiable data on a revoked consent.

11. Patient can run it or take it elsewhereNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

Cloud service licensed to health organizations. No patient route to run it or move the function with their data.

12. No advertising or data businessYesdocumented
Not independently replicated
Microsoft will not disclose or provide access to any Processed Data except: (1) as Customer directs Source

Source is the Microsoft Product and Services DPA (WW, English, May 2026 docx) linked from the Licensing page. The DPA also says Microsoft will not use Customer Data for user profiling or advertising or similar commercial purposes. Caveats: Microsoft uses anonymized data internally to improve models (counted under Q10), uses subprocessors and human annotators, and the DPA binds the customer relationship, not the patient directly.

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

Microsoft Dragon Copilot: CAIHL reassessment

September 8, 2026. Published AI-assisted draft.

Microsoft Dragon Copilot supports clinician-controlled documentation and nursing workflows. Clinicians approve the output. Declining encounter recording may still allow later AI processing of clinician dictation, and revocation does not remove data already anonymized.

Scope and agency posture

Microsoft Dragon Copilot clinician documentation and nursing workflows, deployed by healthcare organizations. Institutional AI affecting the patient record, not a patient-operated assistant.

Posture: Mixed, institution-led.

Axis: 36/100, retained provisionally. No numerical recalibration was performed.

Sources and documented findings

  • The privacy whitepaper, updated August 20, assigns review, editing, transfer, and sign-off to clinicians. It describes post-encounter dictation when a patient prefers not to record.
  • That whitepaper describes anonymization within 90 days and customer-mediated revocation. Already anonymized data is not purged on revocation.
  • The FAQ describes retained audio, text, and generated flowsheet data. Public article content was retrievable despite an authorization notice. No login was used.

Patient authority

Inference from the documented workflow: The clinician sets the documentation task and approves its consequences. Refusing encounter recording can change capture method, but is not the same as preventing later AI processing of clinician dictation.

Critical capacity

Editorial assessment: Better documentation may support subsequent patient review, but the product assigns the patient no direct evidence-inspection or editing workflow in these materials. Clinician benefit is not evidence of increased patient authorship.

Informed control

Editorial assessment: The lifecycle and revocation limits are specifically disclosed. Effective consent still depends on what the healthcare organization tells the patient and how requests are handled.

Assessment and limits

Mixed, institution-led remains supported by explicit allocation of control, rather than institutional ownership alone. The update adds a concrete distinction between recording refusal and AI-use refusal.

Confidence: Medium for product governance.

Remaining uncertainty: Local consent wording, access to transcripts, AI-specific correction routes, patient-facing release of outputs, and actual enforcement.

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.