Software as a Medical Device (SaMD) is moving faster than many traditional medical device regulatory models were designed to accommodate. Artificial intelligence (AI), machine learning (ML), adaptive algorithms and continuous software updates create questions that extend beyond initial product approval: How should risk be categorized? What evidence demonstrates performance? How should algorithm changes be controlled? And how can regulators avoid completely different requirements for the same software across markets?
Taiwan’s Food and Drug Administration (TFDA) has become increasingly active in answering these questions. In February 2026, TFDA was re-elected as Chair of the Global Harmonization Working Party (GHWP) Working Group 3 on Pre-market: Software as a Medical Device, placing Taiwan in an important position within international discussions on SaMD and AI-enabled medical devices.
For device manufacturers and sponsors, understanding the role of TFDA in SaMD Harmonization is therefore about more than preparing a Taiwan submission. It provides insight into how Taiwan’s regulatory framework is evolving alongside broader international principles for software risk, lifecycle management, clinical evidence and AI governance.
Why Is TFDA Important to Global SaMD Harmonization?
SaMD can be developed in one country, hosted in another and deployed across multiple healthcare systems without any physical device crossing a border. Regulation, however, remains jurisdiction specific.
That creates a practical challenge for developers. A software product may face different expectations for:
- Risk classification
- Software documentation
- Clinical and analytical performance
- Cybersecurity
- AI/ML training and validation datasets
- Quality management
- Software changes and updates
- Post-market performance monitoring
Global SaMD Harmonization aims to reduce unnecessary regulatory fragmentation by developing shared terminology and regulatory principles while allowing individual authorities to retain their own legal requirements.
TFDA’s current international role is particularly relevant here. As Chair of GHWP WG3, TFDA has identified priorities that include regulatory frameworks for emerging SaMD technologies, algorithm transparency, cybersecurity requirements, international regulatory collaboration and capacity building. TFDA also reports that the working group has been addressing areas such as definitions, risk categorization and lifecycle management for AI-driven SaMD.
This does not create a single worldwide approval pathway. Instead, harmonization helps regulators and manufacturers work from increasingly comparable regulatory concepts.
How International SaMD Principles Are Shaping Taiwan's Approach
One of the most influential global frameworks comes from the International Medical Device Regulators Forum (IMDRF).
IMDRF defines SaMD as software intended for one or more medical purposes that performs those purposes without being part of a hardware medical device. Its work has established internationally referenced concepts covering SaMD definitions, risk categorization, quality management and clinical evaluation.
More recently, IMDRF published its Good Machine Learning Practice for Medical Device Development: Guiding Principles in January 2025.
Taiwan’s harmonization efforts are also reflected in its domestic classification framework. In August 2023, TFDA revised its Medical Device Classification Regulations, adding five SaMD product items and adjusting existing classification scopes. According to TFDA, the changes were intended to bring Taiwan’s SaMD classification model closer to international regulatory norms while reflecting local device-management requirements. This makes classification an important early step for developers planning a Taiwan SaMD submission.
In June 2026, TFDA published its own principles for Good Machine Learning Practice (GMLP) for AI medical devices, explicitly referencing the IMDRF GMLP principles. The Taiwan framework addresses areas including product design, software engineering quality, representative datasets, algorithm-related risks and post-market performance monitoring throughout the product lifecycle.
This is one of the clearest examples of how global regulatory convergence can influence practical national requirements without replacing the local regulatory system.
What Does TFDA's Current SaMD and AI/ML Guidance Cover?
Current TFDA SaMD Guidance spans several technical and lifecycle areas rather than a single regulatory document. TFDA SaMD Guidance has developed progressively as software and AI technologies have become more sophisticated.
AI/ML Medical Device Software
TFDA maintains technical guidance for medical device software using AI/ML technologies. For registration, applicants may need to describe the software functionality and architecture, indicate whether the algorithm is locked or adaptive, and define relevant performance targets based on intended use and product claims.
This means a SaMD submission cannot rely on a general description of an algorithm. Sponsors need to connect the algorithm to its intended medical purpose, technical architecture, performance characteristics and risk profile.
CADe and CADx Software
In August 2025, TFDA revised its guidance covering AI/ML-based computer-assisted detection (CADe) and computer-assisted diagnosis (CADx) medical devices.
The guidance applies to relevant Class II and III products and includes standalone SaMD as well as software incorporated into other devices. TFDA identifies technical documentation expectations around software architecture, algorithm design, intended purpose, performance, warnings and validation.
AI/ML Predetermined Change Control Plans
A major challenge with AI-based SaMD is that software may need to change after approval.
In September 2024, TFDA issued guidance for Predetermined Change Control Plans (PCCPs) for AI/ML medical device software. PCCPs are becoming increasingly important internationally because a purely static approval model may not be suitable for software whose algorithms, datasets or performance characteristics evolve.
Good Machine Learning Practice
TFDA’s 2026 GMLP principles extend the regulatory discussion beyond submission documentation.
They promote lifecycle thinking around:
- Intended use
- Multidisciplinary expertise
- Software engineering quality
- Training and testing data
- Algorithm bias
- Data drift
- Clinical performance
- Real-world monitoring
This makes the regulatory strategy relevant from product development through post-market use rather than only at the point of registration.
Key Areas Where TFDA and Global SaMD Principles Converge
Several themes now appear consistently across Taiwan and international SaMD frameworks.
For global developers, these common concepts create an opportunity: build the core regulatory evidence package around internationally recognized principles first, then identify the additional requirements that apply in each country.
What Does Harmonization Mean for Device Sponsors Entering Taiwan?
Harmonization can reduce duplication, but it does not mean that an FDA, EU or other international authorization automatically provides Taiwanese market access.
Taiwan maintains its own requirements under the Medical Devices Act and associated regulations. Depending on device classification, applicable listing or registration and market-approval requirements must still be addressed. TFDA’s current regulations separately govern medical device classification, licensing, quality management systems and market authorization.
For imported products, foreign manufacturers also need to consider Taiwan-specific representation and documentation. TFDA requires an authorization arrangement identifying the Taiwan agent responsible for registration and market approval of imported medical devices.
Quality system requirements must also be assessed. TFDA states that medical device manufacturers are subject to Taiwan’s Medical Device Quality Management System framework, while foreign manufacturers may need the applicable QSD conformity assessment before market entry.
Chinese labelling and instructions are another local requirement that should not be overlooked. TFDA requires manufacturers or importers to provide Chinese labelling and instructions for applicable medical devices before sale, subject to specified exceptions.
A Practical Taiwan SaMD Readiness Approach
Before submitting a SaMD product in Taiwan, sponsors should:
- Confirm whether the software qualifies as a medical device based on its intended medical purpose.
- Determine its Taiwan classification and regulatory pathway.
- Map existing global evidence against TFDA requirements rather than rebuilding the entire submission from the beginning.
- Review software architecture, performance and clinical evidence for consistency with current TFDA SaMD guidance.
- Assess AI/ML-specific requirements, including dataset suitability, algorithm validation and GMLP considerations where relevant.
- Plan software modifications early, including whether a PCCP approach is appropriate.
- Confirm Taiwan QMS requirements and applicable QSD conformity assessment for foreign manufacturing sites.
This approach captures the benefit of harmonization while recognizing that market access remains locally regulated.
Why Lifecycle Planning Matters More for AI-Enabled SaMD
Traditional medical devices may remain materially unchanged for long periods. Software does not.
SaMD developers frequently need to address:
- Operating-system updates
- Cybersecurity patches
- Algorithm refinements
- New datasets
- Changes in clinical workflow
- Model retraining
- New functionality
- Performance drift
The regulatory strategy should therefore distinguish between routine software maintenance and changes that may affect safety, performance or intended use.
TFDA’s PCCP work and its 2026 GMLP principles reflect this shift toward lifecycle-based regulation. Internationally, IMDRF is also continuing work on software-specific risk, AI lifecycle management and predetermined change-control concepts.
How Can DDReg Help?
DDReg provides Regulatory Services in Taiwan to support medical device and SaMD companies across Taiwan-specific regulatory requirements and broader global development strategies.
Support can include:
- SaMD regulatory pathway and classification assessment
- TFDA regulatory strategy
- Taiwan market-entry planning
- Technical documentation gap assessment
- AI/ML and software documentation review
- Clinical evidence and performance documentation support
- QMS compliance and QSD conformity-assessment readiness for foreign manufacturers, where applicable
- Taiwan authorized-agent coordination
- TFDA product registration support
- Labelling and IFU compliance
- Regulatory change assessment
- Post-market and lifecycle management support.
For companies developing SaMD across multiple markets, DDReg can also help identify which technical and clinical evidence can form part of a common global regulatory package and where Taiwan-specific adaptations are still required. DDReg’s Taiwan regulatory services cover medical device registration, technical documentation, local representation, TFDA coordination and ongoing lifecycle compliance.
Conclusion
TFDA’s growing role in Global SaMD Harmonization reflects a broader change in medical device regulation: software can no longer be regulated effectively through a one-time, market-by-market approval mindset alone.
Through its leadership of GHWP WG3, evolving TFDA SaMD Guidance, AI/ML change-control framework and 2026 Good Machine Learning Practice principles, Taiwan is contributing to greater international convergence around how SaMD should be developed, evaluated and managed throughout its lifecycle.
For device sponsors, harmonization creates an opportunity to design regulatory evidence that can support several markets. It does not remove Taiwan-specific requirements. Successful market entry still depends on correctly aligning global technical evidence with TFDA classification, registration, quality-system, local representation, labelling and lifecycle expectations.
Building that alignment early can reduce regulatory rework and create a stronger foundation for both Taiwan market access and broader global SaMD development.
Frequently Asked Questions
TFDA currently chairs GHWP Working Group 3 on Pre-market SaMD. The group works on regulatory convergence in areas including SaMD definitions, risk categorization, AI/ML lifecycle considerations, transparency and cybersecurity.
Taiwan maintains its own legally applicable regulatory framework, but TFDA uses international regulatory principles as important references. Its 2026 GMLP principles, for example, explicitly reference IMDRF's Good Machine Learning Practice principles.
No. Harmonization develops greater consistency in regulatory principles, but individual jurisdictions retain their own classification, registration, quality, representation and post-market requirements.
Depending on the product, documentation may need to address software architecture, intended purpose, algorithm type, performance specifications, validation, clinical evidence, risk management and applicable AI/ML-specific considerations.
Yes. TFDA has published guidance relating to Predetermined Change Control Plans for AI/ML medical device software, supporting structured planning for certain anticipated software modifications.