Software as a Medical Device and AI Health Tools
Software as a medical device is regulated by Health Canada under the Medical Devices Regulations when its intended use is medical. Class II to IV products need a licence, AI tools face specific evidence expectations, and predetermined change control plans can allow model updates without fresh amendments.

Key takeaways
- Intended use, shown by labelling and marketing, decides whether software is a medical device.
- Class II, III and IV devices need a medical device licence before import or sale.
- Health Canada's machine learning guidance expects representative data and subgroup performance reporting.
- An authorized PCCP lets planned model changes proceed without a significant-change amendment.
- Serious incidents must be reported within 10 or 30 days, with periodic summary reports.
Software as a medical device in Canada: the short answer
Software as a medical device (SaMD) is regulated in Canada like any other medical device when it is intended to diagnose, treat, mitigate or prevent disease, or to monitor or support those purposes. Health Canada applies the Food and Drugs Act and the Medical Devices Regulations (MDR): the software is classified from Class I (lowest risk) to Class IV (highest), Class II to IV products need a medical device licence before they are sold, and manufacturers carry continuing post-market duties.
For artificial intelligence, Health Canada's guidance on machine-learning-enabled medical devices (MLMD) adds expectations on data, performance and transparency, and allows a predetermined change control plan (PCCP) so that certain model updates can be made without a fresh licence amendment.

For developers of clinical decision support, diagnostic imaging AI, digital therapeutics and remote monitoring tools, the key legal questions are whether the product is a device at all, what class it falls into, what evidence Health Canada will expect, how updates to a learning model are handled, and what must be reported after launch. This article covers each, along with privacy and procurement issues that shape adoption by hospitals and provinces.
Is the software a medical device?
The starting point is intended use. Software is a medical device if the manufacturer intends it to be used for a medical purpose such as diagnosis, treatment, mitigation or prevention of a disease or abnormal physical state, or for related monitoring. Marketing claims, labelling and instructions for use all evidence intended use, so a product described as "AI that detects early sepsis" is likely a device even if its developer sees it as an analytics tool.
Health Canada's SaMD guidance distinguishes software that is excluded from device regulation, such as administrative tools, general wellness apps and some clinical decision support that only supports a health care professional who can independently review the basis for a recommendation. Software that analyzes medical images, physiological signals or patient-specific data to produce a diagnosis or treatment recommendation that the clinician cannot readily verify is generally a device. The line is fact-specific, and many products contain both regulated and unregulated modules.
Write the intended use statement before writing marketing copy. Health Canada will read websites, investor decks and sales materials as evidence of intended use, and a claim made casually in a sales presentation can turn an unregulated tool into a medical device or move it into a higher class.
Classifying software as a medical device
Classification follows the rules in Schedule 1 of the MDR, interpreted for software through Health Canada's SaMD classification guidance. In practice, the significance of the information the software provides (to treat or diagnose, to drive clinical management, or to inform it) and the seriousness of the condition (critical, serious or non-serious) determine the class.
| Class | Typical SaMD examples (illustrative) | Market authorization |
|---|---|---|
| Class I | Low-risk software, such as some tools that only inform management of non-serious conditions | No device licence; importers and distributors generally need an establishment licence |
| Class II | Software that informs or drives management of non-serious or serious conditions | Medical device licence with declaration-based evidence and a recognized quality management system certificate |
| Class III | Software that diagnoses or drives treatment of serious conditions, such as many imaging AI tools | Licence with summary of safety and effectiveness evidence and clinical data |
| Class IV | Software that diagnoses or treats critical conditions | Licence with full review of safety and effectiveness evidence |
Under s. 26 MDR, no one may import or sell a Class II, III or IV device unless the manufacturer holds a licence for it, or an amended licence where the device has undergone a significant change described in s. 34. Section 27 prohibits advertising those devices for sale unless the manufacturer holds a licence, subject to limited exceptions. Under s. 44, importers and distributors generally need a medical device establishment licence, although manufacturers of licensed Class II to IV devices are exempt for their own products.
Pre-market evidence for AI health tools
A licence application for software as a medical device must demonstrate that the product meets the safety and effectiveness requirements in ss. 10 to 20 MDR and the labelling requirements in ss. 21 to 23. For SaMD, the evidence typically includes software development life cycle documentation, cybersecurity controls, verification and validation testing, usability engineering, and clinical evidence proportionate to the claims and class.
Health Canada's pre-market guidance for machine-learning-enabled medical devices (published 1 April 2026, page modified 17 September 2026) covers applications for Class II, III and IV MLMD. It expects manufacturers to describe the model and its intended use, explain how training, tuning and test data were selected and kept independent, show that data are representative of the intended Canadian population, report performance with appropriate metrics across relevant subgroups, and address risks such as bias, data drift and automation bias.
The guidance references the internationally developed good machine learning practice principles. The guidance also notes that the Minister may impose terms and conditions on a licence under s. 36(2) MDR to manage uncertainties about benefits and risks.
Predetermined change control plans
Machine learning models are often retrained or tuned after launch. Under s. 34 MDR, certain changes to a licensed device, including significant changes to Class III and IV devices, require a licence amendment before the changed device can be sold. That is unworkable for frequent model updates. Health Canada's guidance addresses this through the PCCP: documentation, submitted with a new licence application or an amendment, that describes the planned changes, the protocol for developing, validating and implementing them, and an impact assessment.
Changes made in accordance with an authorized PCCP do not need a separate significant-change amendment, although they must be documented in the quality management system. A PCCP cannot be used to change the device's medical conditions, purposes or uses; those still require a new or amended licence. Typical PCCP content includes retraining with new data to maintain or improve performance, adjustments to operating thresholds, and expansion to new input devices of the same type, each with acceptance criteria.
Post-market duties for software as a medical device
Licensing is only the beginning. Manufacturers and importers must:
- report incidents in Canada involving a device failure, deterioration in effectiveness or inadequate labelling that led, or could lead, to death or serious deterioration in health (s. 59), with preliminary reports due within 10 or 30 days depending on severity (s. 60);
- prepare periodic summary reports analyzing adverse effects and problems, biennially for Class II and annually for Class III and IV devices (s. 61.4);
- maintain distribution records and complaint handling procedures, and carry out and report recalls (ss. 63-65); and
- respond to Health Canada requests for information, assessments and tests, and to orders made under the Food and Drugs Act.
For SaMD, post-market monitoring should include real-world performance tracking, drift detection, cybersecurity vulnerability management and processes for rolling back faulty updates. Software defects that affect clinical outputs can trigger both incident reporting and recall obligations.
Cybersecurity for software as a medical device
Cybersecurity is part of safety and effectiveness for software as a medical device. Health Canada expects manufacturers of connected devices and software to address cybersecurity throughout the product life cycle: threat modelling and secure design during development, testing before release, documentation of third-party and open-source components, and a plan for monitoring vulnerabilities and deploying patches after launch. A licence application that does not explain how the software will be kept secure over its service life is likely to attract questions.
Cloud-hosted software as a medical device raises additional questions about where data and models are hosted, who controls updates, and how the manufacturer ensures that the version used by each customer is the licensed version. Hospital customers will often require security assessments, penetration test results and incident notification commitments in contracts, independently of Health Canada's requirements.
Privacy, procurement and liability
Health AI depends on data, and privacy law shapes both development and deployment. Hospitals and clinicians are generally governed by provincial health information statutes, such as Ontario's Personal Health Information Protection Act, while private developers are governed by PIPEDA or the private-sector laws of Quebec, British Columbia and Alberta. Training a model on patient data typically requires a lawful basis such as de-identification, research ethics approval or consent, and data-sharing agreements with institutions. Quebec requires privacy impact assessments for transfers of personal information outside the province. See our page on privacy and data protection.
Adoption usually runs through provincial health systems and hospital procurement, which bring their own evaluation, cybersecurity and contracting requirements. Contracts should address liability for clinical outputs, indemnities, data rights for improving the model, service levels, and what happens if Health Canada suspends the licence. Professional liability remains with clinicians, but manufacturers face product liability claims for defective software and inadequate warnings. Canada does not yet have a general AI statute in force; see our page on artificial intelligence.
A launch roadmap for SaMD developers
- Define intended use and claims, and decide whether the product is a device.
- Classify under Schedule 1 using Health Canada's SaMD guidance, and identify any modules outside device regulation.
- Establish a quality management system suitable for certification, with software life cycle and cybersecurity processes.
- Plan clinical and performance evidence that reflects the Canadian population, with subgroup analysis.
- For machine learning, decide whether to include a PCCP in the first application.
- Build post-market surveillance, incident reporting and recall procedures before launch.
- Prepare privacy, data-sharing and procurement documentation for hospital customers.
Timing matters for financing as well as compliance. Investors in digital health companies typically ask for a regulatory roadmap that links product milestones to classification, evidence generation and submission dates, and they will discount revenue forecasts that assume sales before a licence is in hand. Early pre-submission engagement with Health Canada, where available, can reduce uncertainty about classification and evidence, and should be documented carefully so that the advice can be relied on in later discussions with the regulator and with investors.
Developers already authorized in the United States or Europe can reuse much of their evidence, but they should not assume that a foreign authorization or classification transfers automatically to Canada.
How TRW & Co helps
We advise developers, investors and health systems on software as a medical device and AI health tools in Canada. We help determine device status and classification, plan licence and amendment strategies including PCCPs, review labelling and marketing claims, draft data-sharing, licensing and hospital procurement agreements, and design post-market compliance processes. We also advise on financings and acquisitions of digital health companies. Regulatory submissions are prepared with the client's regulatory and quality teams. Our work connects with our life sciences and health sector team; see also our insight on freedom of information requests for protecting regulatory filings. To discuss a product, contact us. This article is general information, not legal advice.
Frequently asked questions
When is health software regulated as a medical device in Canada?
When its intended use is medical, such as diagnosing, treating, mitigating or preventing disease or monitoring a condition. Health Canada looks at labelling, instructions and marketing claims. Administrative and general wellness software is generally outside device regulation, and some clinical decision support may be excluded.
What is a predetermined change control plan?
A PCCP is documentation submitted with a licence application or amendment for a machine-learning-enabled device. It describes planned model changes, the protocol for developing and validating them, and an impact assessment. Changes made under an authorized PCCP do not need a separate significant-change amendment, but cannot change the device's intended use.
Do Class I software devices need a licence?
Class I devices do not need a medical device licence, but importers and distributors generally need a medical device establishment licence, and manufacturers must still meet safety, labelling and post-market obligations such as complaint handling and recalls.
What must be reported after a SaMD is launched?
Manufacturers and importers must report serious incidents in Canada to Health Canada, with preliminary reports within 10 or 30 days depending on severity, prepare periodic summary reports, maintain complaint and distribution records and carry out recalls when required.
Can a US FDA clearance be used in Canada?
Evidence prepared for the FDA or European regulators can often be reused, but Canada has its own classification rules and licensing process. Developers should confirm the Canadian class, prepare a Canadian application and check that clinical evidence reflects the Canadian intended population.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


