Artificial intelligence–powered medical scribe tools (AI scribes) are increasingly used by Canadian health organizations to reduce documentation workload. In response, privacy regulators in Ontario, British Columbia, and Alberta have issued guidance on their lawful use under provincial health-privacy laws.
The scope of the new guidance varies by province. In Ontario and Alberta, it applies broadly to health information custodians as defined under PHIPA and the HIA, including hospitals, clinics, and physicians. In British Columbia, it applies only to private-sector healthcare providers regulated by PIPA, such as independent practitioners and most primary care clinics, and does not apply to public bodies like hospitals or health authorities.
This article summarizes the core compliance requirements in each province and highlights practical steps organizations should take before implementing AI scribe solutions.
Key takeaways for health organizations
As regulatory oversight of AI in healthcare increases, health organizations operating in Ontario, British Columbia, and Alberta should ensure that their use of AI scribes aligns with applicable legal requirements. In practical terms, this includes:
- Documenting lawful authority and necessity, including the justification for using AI scribes despite their privacy impact.
- Defining the scope of authorized use, including whether the scribe will also be used to support medical decision-making and clinical workflows.
- Planning how to obtain consent (which is required in all jurisdictions).
- Prohibiting secondary use (including vendor use of inputs and outputs for training) through contracts and system controls.
- Limiting retention outside of the Electronic Medical Record (EMR) and relying on summary-only records.
- Requiring human review of all AI-generated content before or soon after entry into the EMR.
- Maintaining strong vendor oversight and governance, supported by a Privacy Impact Assessment (PIA) and, if a scribe will also be used to support medical decision-making, an Algorithmic Impact Assessment (AIA).
What is an AI scribe?
AI scribes are tools that use generative AI to capture clinical conversations, usually through real‑time audio, and turn them into structured notes and summaries. They can ease documentation pressures and help clinicians stay focused on patient interactions during care delivery. Because these tools record entire conversations involving patients, clinicians, and others, they typically process sensitive personal health information. Some AI scribes now go beyond simple transcription to provide features that support medical decision-making and clinical workflows. The deployment of scribes can raise documentation and other clinical risks and engages compliance obligations under provincial health-privacy laws.
AI scribe adoption must be “necessary” and “reasonable”
Across all three provinces, regulators emphasize that AI scribes cannot be adopted for convenience alone. Organizations must satisfy the applicable legal thresholds and justify the scope of recording full clinical encounters, most notably by showing that such collection is necessary for their operations and appropriate given the sensitivity of the information involved.
- Ontario: PHIPA’s data-minimization rules require custodians to collect and use only the personal health information needed for the intended purpose. For AI scribes, this means limiting what is recorded, ensuring there is legal authority for the collection, and using less intrusive options where possible.
- British Columbia: Under PIPA, organizations must meet the “reasonable person” test by demonstrating that a reasonable person would consider the collection of audio recordings, including sensitive voice data and incidental third-party information, appropriate in the circumstances, and that no less privacy-intrusive alternative would achieve the same objective.
- Alberta: HIA applies the strictest standard. Custodians may collect personal health information only when expressly authorized by law, regardless of consent. Recording full clinical encounters must therefore be essential for care or another permitted purpose (e.g., determining eligibility or supporting internal management), not simply efficient.
Given the utility of AI scribes, we do not expect privacy regulators to challenge their use, but justification is nonetheless important to document, and custodians and trustees should carefully decide upon the scope of their use.
Consent must be jurisdiction-specific and legally grounded
Patient consent for the use of AI scribes has quickly become the norm. Consent is required in all three jurisdictions.
- Ontario: Although health information custodians do not normally obtain consent to use technologies in processing personal health information, the IPC/Ontario has said that “consent of individuals would generally be required." According to the IPC, patients must understand that the encounter will be recorded using AI, what information is collected, the involvement of vendors, and the key risks and benefits of using AI scribes.
- British Columbia: The OIPC has said that BC healthcare organizations must obtain consent before using an AI scribe in almost all clinical situations. Patients must be told what the AI scribe is, what it records, how their information (and any third‑party information) will be used, and the associated privacy risks, and must have a real choice to accept or decline.
- Alberta: In Alberta, consent is required based on a provision of the HIA that requires custodians to obtain written consent to collect health information by a device that may not be visible to the individual. While custodians may comply by adopting processes to make the recording devices visible, they are best to obtain consent.
The legal standard for consent varies by jurisdiction. Practically, through, all custodians and trustees should adopt the disclosure called for by the Ontario and British Columbia commissioners. Those in Ontario and British Columbia who elect to obtain consent orally should require their clinicians to document the consent.
Secondary use and vendor training are effectively precluded
All three regulators take a restrictive approach to any secondary use of AI scribe inputs or outputs, particularly for vendor model training.
- Ontario: PHIPA permits secondary use only with valid consent or if personal information is de‑identified to a “very low” re‑identification risk. IPC guidance indicates that audio recordings almost never meet this standard, effectively prohibiting vendor training.
- British Columbia: Under PIPA, secondary use without consent is allowed only when it is reasonable and specifically authorized by law. OIPC guidance indicates that voice data usually remains personal information even when de-identified.
- Alberta: Under HIA, vendors are treated as affiliates and may use personal health information only for purposes authorized by the custodian. Model training using personal health information is not permitted and cannot be justified by consent.
Custodians and trustees should carefully consider the secondary use issue. De-identification is a common means of enabling secondary use of personal health information, but there are strict requirements as to who de-identifies and how, and the very nature of the data produced by AI scribes renders lawful secondary use of very limited possibility.
Recordings and transcripts should be retained only in exceptional cases
Regulators are aligned in treating audio recordings and full transcripts as high-risk due to voice biometrics, incidental capture, and heightened breach exposure. As a result, the guidance converges on a principle of minimal retention.
- Ontario: Retention outside the EMR should be exceptional. Best practice is to delete recordings and transcripts once a human‑validated summary is stored.
- British Columbia: Limited retention is allowed only where needed for clinical decision‑making. Voice recordings should be deleted after transcription unless a clear purpose exists.
- Alberta: Recordings and transcripts should be deleted after transcription and clinician validation, consistent with HIA’s essential‑information standard.
In practice, an EMR-only retention model (i.e., clinician-reviewed notes entered into the EMR, with underlying audio and transcripts deleted) has emerged as the prevailing compliance baseline.
Accuracy risks require human oversight
Across Ontario, British Columbia, and Alberta, regulators highlight similar accuracy risks, and all reject autonomous use of AI scribes. AI-generated notes must always be reviewed by a clinician, who remains legally responsible for the accuracy and completeness of the health record.
- Ontario: Human review is required before any AI-generated content is relied on or entered into the EMR. Policies must be implemented and address hallucinations, multilingual errors, and automation bias.
- British Columbia: PIPA requires organizations to tale reasonable steps to ensure accuracy, including human-in-the-loop review, staff training, and periodic audits.
- Alberta: Custodians must ensure accuracy under HIA. AI outputs must be reviewed to prevent errors such as session mixing or mis‑transcription.
Accountability and vendor oversight cannot be delegated
In all three provinces, accountability remains with the custodian. Vendors may provide AI scribe technology or processing support, but health organizations remain fully responsible for compliance with health-privacy laws.
- Ontario: The IPC expects strong governance, including PIAs/AIAs as appropriate, clear policies, and detailed vendor contracts covering use restrictions, security, breach reporting, and audit rights.
- British Columbia: Organizations must maintain updated policies, ensure staff training, implement strong security safeguards, and manage access, correction, and complaint processes. Vendor contracts must prohibit unauthorized use and ensure secure handling.
- Alberta: Vendor contracts must preserve custodial control over health information and require vendors to comply with all HIA obligations, including limits on use and secure destruction.
