AI Scribes Demonstrate Healthcare AI’s Potential, but Fragmentation Risks Canada’s Commercialization Opportunity

[Sandra Holdsworth]

August 25, 2026 – Artificial intelligence (AI) scribes have demonstrated strong clinician demand and measurable benefits, but Canada’s fragmented procurement and regulatory systems could prevent the country from safely translating early adoption into a competitive healthcare AI sector, according to a new Communiqué from the C.D. Howe Institute’s Health Sector Economic Growth and Resilience Working Group.

Composed of experts from the public sector, business, and academia, the group met for its fourth session to examine the adoption, oversight, and commercialization of AI in Canada’s healthcare system, with a particular focus on the rapidly scaling use of AI-generated clinical documentation, or “AI scribes.” The meeting coincided with a federal-provincial announcement expanding shared clinical data infrastructure as part of Canada’s National AI Strategy.

The discussion began with a presentation from Shelley Spence, Auditor General of Ontario, warning of the governance and accuracy gaps in the initial procurement of these tools. A recent Ontario Public Service audit found that employees could access unsanctioned and unsecured AI tools from government-issued devices and that very few had completed recommended training. Separately, Supply Ontario’s assessment of AI scribe vendors found that all 20 evaluated products exhibited some incorrect medical information, including AI hallucinations or incomplete documentation. As AI tools continue to rapidly evolve, so must the public systems responsible for evaluation, validation and procurement.

Krista Balenko, Vice President, Enablement and Operations at Canada Health Infoway, then explored the rapid scale in use, citing a national AI Scribe initiative that offered primary care clinicians a 12-month funded licence, with demand far exceeding expectations, as more than 10,000 providers registered within two weeks. An independent evaluation also found that roughly 90 percent of clinicians considered the tools valuable, and nearly 70 percent reported less administrative burden.

Gavin Tong, Canadian Health Industry Data and AI Lead at Accenture, then examined Canada’s healthcare AI commercialization gap relative to international peers. Although Canada has strong AI research, rich longitudinal health data and an educated workforce, companies must still navigate separate procurement processes, regulatory requirements and reimbursement pathways across 13 provincial and territorial jurisdictions. Participants argued that greater coordination could make Canada a far more attractive market for health technology companies, although some cautioned that fragmentation reflects broader features of Canadian health federalism that procurement reform alone cannot solve.

The Working Group concluded that Canada has many of the foundations needed to become a leader in healthcare AI. While questions remain around public trust, AI use outside governed channels and workforce impacts, participants generally viewed the risks of under-investing in healthcare AI as greater than the risks of broader adoption, provided accuracy, bias and security safeguards keep pace with deployment.

Read the Full Report

For more information, contact: Raquel Schneider, Communications Officer, C.D. Howe Institute, 647-805-3918, rschneider@cdhowe.org.

The C.D. Howe Institute is an independent not-for-profit research institute whose mission is to raise living standards by fostering economically sound public policies. Widely considered to be Canada’s most influential think tank, the Institute is a trusted source of essential policy intelligence, distinguished by research that is nonpartisan, evidence-based and subject to definitive expert review.

Want more insights like this? Subscribe to our newsletter for the latest research and expert commentary.

[Debra Turnbull]

@Sandra-Holdsworth @Alies-Maybee

After reading this a few times – it finally hit me: where's the Patients' Voice in all of this? Yes, there is representation from the public sector, business, and academia... but how about un-biased patients?

AI Scribes hallucinate, capture inaccurate and erroneous information... yes, we know that already. However, it's nice to see that Ontario's Auditor General confirms this in her audit report (found here). Clinicians find AI Scribes to be useful in the alleviation of administrative burden... but what about the errors? How do we protect ourselves?... Simple: ask for a copy of the transcript, or the EMR notes at your next visit. I did. I now have a printout of my visit so that I can refresh my memory – before my next appointment. You can ask too.

The focus of the Howe report is on Canada's commercialization of its healthcare AI tools. The problem is that training AI needs vast amounts of data – our health data. Your Personal Health Information (PHI) is locked away in the EMR silos of your doctors' offices, clinics, specialists, hospitals, etc. Here's the thing – did you give your CONSENT for commercial use? I know I didn't.

The federal government has a project that could establish a consent mechanism - by first establishing a standard. A Canadian Standard for Data Consent Management – CAN/DGSI 133 - is now open for public review. That means you can leave a comment on this Standard – by directly inserting it in the standard document. I have left some; including an essay in Annex A.

Much like the Standard for AI in CDS [Chris' post], two unique components that will be needed in any AI product: Trust and Explainability. Trust (or public trust) gets established when the AI product shows itself to be safe, and conforms with Canadian privacy rules. Explainability is a feature of the AI tool whereby it can produce an output explanation of how it came to its decisions.
What does

Canada does have AI infrastructure. The problem is that it's missing the Big Data component. In the healthcare space – that's our Personal Health Information, and NOT personal information as the Consent Standard proposes. (Personal Information as DGSI 133 mentions, are those pieces of information such as address or DoB that identify you – in the commerce ecosystem. Personal Identifiers are those info pieces that ID you in the healthcare system.) It will be upon the commercial enterprises to ask, and explain (in a manner that we understand) how it will use your data. You have the right to say: “no”.

What does this all mean
C.D. Howe missed the mark. The commercialization of healthcare AI in this country can not happen without us – the patient!! They want our data – it's valuable – yet they don't want to tell us how it will be used. I continue to see this bias in: research projects, high tech circles, hospital projects, etc. They are oblivious to the realization that patients are the solution to their problems. Oh well...

So, to recap:
If your doctor uses an AI Scribe – ask for a printout: transcript or notes... and don't forget to take our AI Scribes survey;
read the standards and add comments to:

• CAN/DGSI 133 Data Consent Management: Consent-driven Approach to Data Access, Interoperability and Mobility , and

• CAN/DGSI 138 Patient-Visible and Accountable AI in Clinical Decision Support. .

(Apollogees if the links are not the same throughout - was trying to provide as much reference info as possible.)