From triaging imaging scans to predicting patient deterioration, AI‑enabled medical devices are now central to clinical decision making in ways unthinkable even a decade ago. Yet behind every high‑performing AI model lies an invisible foundation: data.

As the healthcare industry accelerates toward AI‑integrated care, one truth has become impossible to ignore: the safety, reliability and accountability of AI systems depend entirely on the quality of the data used to build them. And until now, the medical device sector has lacked a unified, lifecycle‑driven framework for managing that data.

PD IEC PAS 63621:2026 changes that.

Developed by IEC Technical Committee 62 with input from relevant stakeholders, this new Publicly Available Specification (PAS) offers a high‑level framework for data management in AI‑enabled medical devices. It supports safer, more trustworthy AI at a time when global health systems need it most.

In this article, we explore the significance of the PAS, its role in the rapidly evolving AI landscape and why organizations across the medical device ecosystem should be paying attention.

The healthcare AI boom: Promise, pressure and a new kind of risk

AI has rapidly shifted from experimental technology to everyday medical infrastructure.

Hospitals now rely on AI to ease radiology backlogs, streamline pathology workflows, improve bed management and tailor treatment decisions. From wearables to image‑guided surgical systems, machine learning is becoming integral to medical device innovation.

But as adoption accelerates, a new regulatory and ethical challenge has emerged: AI is only as dependable as the data behind it. Inadequate or poorly controlled datasets can introduce bias, cause performance drift, compromise safety, obscure clinical evidence and ultimately lead to regulatory non‑compliance.

Global regulators are increasingly demanding rigorous, auditable, lifecycle‑based controls for AI‑related data. Yet until recently, the medical device industry lacked a unified reference for managing such data. Manufacturers created their own processes, external data providers followed different standards and clinical partners operated with limited guidance, resulting in fragmented, inconsistent and often non‑aligned practices.

PD IEC PAS 63621:2026 delivers an internationally aligned framework that supports data management into medical device quality systems.

What PD IEC PAS 63621 sets out to achieve

At its heart, the PAS provides a structured blueprint for managing data across every stage of the AI lifecycle, from initial planning to end‑of‑life decommissioning. It is intentionally high‑level structured, ensuring it applies across all device types, medical specialties and AI architectures.

The PAS recognizes that organizations may be involved in any combination of activities - design and development, production, implementation, servicing, external data supply - and defines requirements flexible enough to fit each scenario.

Its core goals include:

• Ensuring data used in AI devices is safe, reliable and fit for purpose.

• Providing structured processes for dataset creation, validation and ongoing quality improvement.

• Creating common expectations for manufacturers, suppliers, clinical data providers and auditors.

• Embedding data management supporting integration with a BS EN ISO 13485 quality management system.

• Strengthening links between data quality and risk management, particularly for safety‑critical AI.

• Supporting regulatory readiness and reducing ambiguity in data generation.

By offering a complete end‑to‑end lifecycle model, the PAS positions data management not as a technical afterthought, but as a central pillar of product safety.

What PD IEC PAS 63621:2026 covers: A closer look at the lifecycle framework

PD IEC PAS 63621:2026 takes a holistic view of data management for AI‑enabled medical devices, outlining the key stages organizations typically consider when handling data throughout its lifecycle. Rather than prescribing detailed requirements, it provides a structured, high‑level framework that helps align data‑related activities with good practice in the medical‑device context.

The PAS highlights themes such as:

1. Establishing a structured approach to data planning

It encourages organizations to think proactively about the types of data they work with, how that data relates to the intended purpose of the AI system and how it will be documented and managed over time.

2. Considering how data is sourced and selected

The framework reinforces the importance of approaching data acquisition thoughtfully, taking into account factors like suitability, diversity, privacy and the broader context in which data originates.

3. Supporting robust data development and annotation practices

It touches on the need for clarity and consistency in how data is described, labelled, organized and updated. This includes maintaining transparency about data origins, characteristics, and any transformations applied.

4. Maintaining an ongoing view of data quality

The PAS promotes a lifecycle perspective in which data quality is monitored over time, reflecting factors such as accuracy, relevance, completeness and how those aspects may evolve.

5. Encouraging evaluation and awareness of potential bias

It recognizes the importance of examining datasets for issues that could impact fairness, safety, or performance and of fostering practices that support more reliable and equitable AI outcomes.

6. Managing how data is shared, accessed and stored

Topics such as secure handling, appropriate access, traceability and clear governance structures are highlighted as important to the overall integrity of data processes.

7. Taking a lifecycle view of monitoring and eventual retirement

The framework acknowledges that data changes over time and may need to be reviewed, updated, or responsibly decommissioned as technologies, clinical realities, or organizational needs evolve.

8. Supporting data‑related activities with wider quality management practices

It places data management in the context of established medical‑device quality systems, supporting consistency, traceability and organizational capability.

Taken together, these themes form a cohesive, high‑level reference that helps organizations bring structure, clarity and accountability to the way they manage data for AI‑enabled medical devices, without dictating specific methods or technical solutions.

What makes PD IEC PAS 63621:2026 different?

Beyond its high‑level lifecycle structure, one of the most valuable aspects of PD IEC PAS 63621 is how clearly it reframes data management as a continuous, standards‑aligned responsibility rather than a one‑time technical task.

The PAS is explicitly focused on the data life cycle associated with AI models that form part of a medical device - including data used to train, test and validate models, as well as the data analysed during device operation. In doing so, it directly supports BS EN ISO 13485 quality management systems, providing a structured way to integrate AI‑specific data management into regulated medical‑device processes.

Importantly, PD IEC PAS 63621 is intentionally concise.

Rather than duplicating detailed data‑quality definitions or metrics, it sets out normative requirements for good data management while drawing on a wider ecosystem of international standards for deeper guidance. For example, it requires organisations to define a data quality model based on recognized data‑quality characteristics, pointing users towards established references such as BS ISO/IEC 25012 and the BS EN ISO/IEC 5259 series.

This approach ensures the PAS remains practical and adaptable, while still aligning with globally recognised best practice.

Data quality as an ongoing commitment

A central message of PD IEC PAS 63621 - consistent with many AI‑related standards - is that “deploy and forget” is not an option. Data management, operation and monitoring are product‑lifetime responsibilities that extend right through to decommissioning.

Because the PAS focuses on AI‑enabled medical devices, it highlights two distinct but related data risks:

• the quality and suitability of data used to train and validate AI components; and

• the quality and characteristics of data analysed by those components during real‑world operation.

To address this, the PAS sets out a six‑stage data lifecycle that closely aligns with the framework in BS EN ISO/IEC 5259‑1:

1. Data requirements 2. Data planning 3. Data acquisition 4. Data development 5. Data provisioning 6. Data decommissioning

Crucially, the data development stage is treated as an ongoing cycle of planning, improvement, verification and analysis. This reflects the reality that data quality can change over time - for example as clinical practices evolve or population characteristics shift - and may no longer meet the manufacturer’s requirements or intended use.

The PAS explicitly recognizes data drift as an ongoing risk and points users towards further international standards for more detailed guidance on data‑quality measures and monitoring approaches.

A reminder that AI is not one single technology

Another important contribution of PD IEC PAS 63621 is its reminder that AI is a broad family of technologies, not just large language models or generic machine learning. Different AI approaches behave differently, and those differences have direct implications for data management.

Reflecting this, the PAS introduces additional normative requirements for AI components that use supervised, semi‑supervised or reinforcement learning, encouraging manufacturers to align data‑management choices with the specific characteristics of the AI technology in use.

This reinforces the PAS’s overall message: effective data governance must be risk‑based, context‑specific and informed by technical reality, rather than relying on generic or one‑size‑fits‑all controls.

Practical insight through informative annexes

PD IEC PAS 63621 is supported by three informative annexes that help bridge the gap between principle and practice.

• Annex A expands on data‑management techniques, offering practical suggestions for data improvement activities.

• Annex B includes extracts from a Chinese national standard on datasets for AI medical devices, providing valuable insight into data description, data‑quality requirements and conformity evaluation.

• Annex C illustrates data screening and cleaning through practical examples.

Together, these annexes reinforce the PAS’s role as both a governance framework and a practical reference point for organisations across the medical device ecosystem.

Supporting safer, more transparent AI

Above all, PD IEC PAS 63621 contributes to safer, more predictable AI in clinical environments.

It ensures organizations establish rigorous controls around the data that drives model performance - controls that reduce bias, support fairness, strengthen safety and build trustworthy data.

As AI becomes increasingly integrated into patient care, this level of structure and transparency is essential. The PAS sets the foundation the sector needs today and positions organizations to meet the regulatory and ethical expectations of tomorrow.

Get your copy of PD IEC PAS 63621 here.