Why regulation is becoming AI's greatest enabler

“Don't wait for the deadline to read the rules.”

That's the advice of Dr. Raymond Nistor, neurosurgeon, medical device safety expert, and member of the Bigpicture Advisory Board. Over a career spanning more than four decades, he has worked across clinical practice, industry, and European medical device regulation, including leadership roles at Baxter and an Austrian Notified Body for medical devices and in vitro diagnostics.

Today, Raymond helps organizations navigate one of healthcare's fastest-changing regulatory landscapes. As Europe updates its framework for AI and diagnostics, he believes success will depend on much more than innovative algorithms. Trusted data, robust validation, harmonized standards, and regulatory readiness are becoming just as critical to bringing AI safely into clinical practice. For Bigpicture, that presents a significant opportunity. By combining high-quality pathology data, governance, and technical expertise, the consortium is helping build the foundations needed for trustworthy AI across Europe.

When safety creates a bottleneck

When the In Vitro Diagnostic Regulation (IVDR) was introduced, its ambition was clear: improve the quality and safety of diagnostic devices across Europe. According to Raymond, those objectives were absolutely the right ones. “The intention of IVDR was very noble,” he says. “Criticizing something is easy, but it's important to understand the intention.” The implementation, however, proved far more challenging than expected. Today, pathology AI is regulated as an in vitro diagnostic medical device under the IVDR, bringing specific requirements for classification, performance evaluation, validation, and conformity assessment. While those safeguards are essential, Raymond believes the system has also created practical bottlenecks. “A brilliant algorithm that could catch a rare cancer has to wait 12 to 18 months just for a review slot. That is not a safety feature. That is a capacity problem.” The lesson, he says, is not that regulation should be reduced, but that it should evolve to better support innovation while maintaining patient safety.

One regulation, 27 interpretations

That evolution is already underway. The EU AI Act became generally applicable on 2 August 2026, adding another important regulatory framework for organizations developing and deploying AI in Europe. Its requirements are being introduced in stages, with specific timelines applying to different categories of AI systems.

At the same time, the European Commission's COM(2025) 1023 proposal, published in December 2025, aims to simplify and reduce the regulatory burden surrounding medical devices and in vitro diagnostic medical devices. Together, these developments are reshaping the regulatory environment in which pathology AI will enter clinical practice. Rather than focusing solely on compliance, the conversation is increasingly shifting toward enabling innovation in a safe, practical, and harmonized way.

One challenge, however, remains. “One regulation. Twenty-seven different national interpretations.” For developers and healthcare organizations, that lack of consistency creates uncertainty and delays. Raymond believes greater harmonization, both in regulation and in technical standards, will be essential if Europe wants to accelerate the adoption of AI in healthcare. He also points to the emerging ISO/CD 24051-2 standard as an important step forward. By providing a harmonized framework for digital pathology and AI workflows, the standard could help create a clearer pathway toward regulatory compliance for developers, laboratories, and regulators alike.

A million slides don't make good data

Technology alone is not enough. Developing AI for pathology requires access to large, diverse, and well-governed datasets. But according to Raymond, regulatory-grade data isn't about volume, it's about trust. “Regulatory-grade data is not about volume. A million slides from a single hospital are just one data point at scale. Regulatory grade means trust.”That means data from multiple sites, multiple scanners, annotated by qualified experts and fully traceable from the glass slide to the clinical truth.

Validation follows the same principle. “Most AI validation today is like testing a car on a single road. Good validation means the AI works on every road, in every weather, for every driver.” Only by validating AI across different hospitals, scanners, and patient populations can developers demonstrate that their algorithms are truly ready for clinical use.
 

Building the evidence base regulators can trust

This is where Raymond sees Bigpicture making a real difference. Rather than every organization building its own datasets and benchmarks from scratch, Bigpicture is creating shared infrastructure that can benefit the entire pathology community. “Data without benchmarking is just files. Benchmarking without standards is just numbers.”

By bringing together high-quality data, common standards, governance, and benchmarking, the consortium is creating a trusted environment for developing and evaluating AI. “Bigpicture would become kind of a reference for clinical data.”

Working from shared, trusted datasets could make validation more consistent, improve reproducibility, and simplify how AI solutions are evaluated across Europe. As Raymond explains: “A developer in Vienna and a hospital in Warsaw will work from the exact same trusted evidence base.”

Beyond building one of Europe's largest digital pathology resources, Bigpicture is helping establish the trusted infrastructure needed for AI to move from promising research into everyday clinical practice. As Raymond puts it: “Without regulatory, this big idea is not going to end up at the patient bedside.”