In healthcare AI, it’s not just about innovation anymore. It’s about regulatory clarity. If you’re building an AI health tool without a defined Software as a Medical Device (SaMD) pathway, you’re not making a minor oversight, you’re opening your company up to enforcement actions and getting shut out of health plan coverage. The market is clearly splitting in two, and companies who built their tech with the regulations in mind from day one are pulling way ahead.
The UK’s Medicines and Healthcare products Regulatory Agency (MHRA) offers a really useful, separate perspective on the global AI regulation conversation. For anyone tracking policy, watching what the MHRA does shows you exactly where they align with or break from big frameworks like the EU Medical Device Regulation (MDR). This comparison doesn’t just show the quirks of one regulator. It’s a good signal for where the entire international rulebook is probably headed.
The Value of a National Regulatory Signal
Global harmonization for AI rules is still mostly a dream, so for now, we have to watch individual national regulators to see what’s coming. The MHRA is a perfect example, actively working with the World Health Organization (WHO) and the FDA to align on things like Good Machine Learning Practice (GMLP) MHRA GMLP guidance. Their positions aren’t developed in a vacuum. This collaboration shows they reflect a growing consensus that will shape the bigger rulebooks. For companies like Tempus AI, HeartFlow, and Butterfly Network, reading these signals is just basic strategic planning.
The FDA’s pipeline is clearly flowing, not clogged. We’re talking 1,614 AI/ML device clearances as of September 2026, with 335 of those in 2025 alone and 154 for cardiovascular AI specifically. But look closer. A massive 96% of these are 510(k) clearances, meaning they’re based on older, “predicate” devices, while only 40 went through the tougher De Novo process for new technology. This reliance on predicates gets products out the door fast, but it screams for clearer guidance on truly new AI, and that’s exactly where national regulators are often stepping in first.
Overlaps and Divergences in European AI Regulation
At a high level, the MHRA’s approach looks a lot like the EU MDR. They’re both built on the same foundation of safety, performance, and quality management systems, and both will demand you bring strong clinical evidence and have a plan for post-market surveillance. The EU MDR, for instance, forces any company with an AI device to have a complete Quality Management System (QMS) compliant with ISO 13485 EU MDR QMS requirements. Anyone who has gone for a CE mark knows it’s a much heavier lift than a 510(k) clearance, with its endless documentation and Notified Body audits.
But the two paths really split when you look at the details. The EU AI Act throws in a whole new layer, creating a specific “high-risk” category for healthcare AI with compliance deadlines looming in December 2027 or August 2028. This act piles on obligations for data governance, transparency, human oversight, and cybersecurity that go beyond the MDR. The MHRA cares about these things too, but its specific laws and guidance handle them differently. For a policy analyst, these differences are a goldmine showing where countries are experimenting, but for global companies like Tempus AI, Viz.ai, Paige AI, Aidoc, Butterfly Network, and Nabla, it’s a massive operational headache to manage two different compliance tracks.
The International Regulatory Thread: Tempus AI, HeartFlow, and Butterfly Network
If you want to see what this all means in practice, just look at the regulatory journeys of Tempus AI, HeartFlow, and Butterfly Network. They operate in a world where their AI tools are scrutinized by multiple regulators at once, including the MHRA and the EU’s various Notified Bodies for the MDR. You can see right in their documentation how they’re trying to satisfy this mixed bag of international requirements.
Tempus AI’s work in diagnostics and precision medicine puts them directly in the regulatory hot seat worldwide. HeartFlow had to fight through a maze of different country-specific regulations for its cardiovascular analysis tool, which proves you can’t survive without bulletproof clinical validation. Then there’s Butterfly Network, whose portable AI ultrasound is a classic SaMD case, they have to get both the hardware and the software components through regulatory hoops. When a regulator like the MHRA digs into these companies, it creates a public case study on how specific AI functions like algorithmic bias and data privacy will be judged.
Here’s a number that should get your attention: the FDA reports that 10.3% of cleared AI/ML medical devices end up getting recalled. That rate is way higher than for non-AI devices, and it’s a stark reminder that getting initial clearance is just the first hurdle, not the finish line. Regulators and patient safety advocates are all pointing to that 10.3% statistic as Exhibit A for why we need continuous oversight and rules that cover an AI tool’s entire lifespan.
The Instructive Gap: What Divergence Reveals
The most useful thing you can do is watch a single regulator like the MHRA and compare it to a big framework like the EU MDR to spot the differences. Where the UK and European approaches agree, you can bet that’s becoming an international standard. For instance, the shared emphasis on Good Machine Learning Practice (GMLP) from the FDA, WHO, and MHRA means GMLP is becoming the de facto global baseline for developing these AI/ML medical devices FDA GMLP guidance.
Where the frameworks don’t agree, however, you find a gap, and that gap is a valuable data point. It flags the exact areas where regulatory thinking is still evolving and where different jurisdictions are running their own experiments based on national priorities. For policy analysts, these divergences are a roadmap to future regulatory trends and strategic opportunities for companies planning their market entry. Understanding these differences lets you make a much more nuanced prediction of how the international regulatory field for AI in healthcare will eventually coalesce.
Frequently Asked Questions
How does the MHRA’s approach to AI regulation compare with the EU Medical Device Regulation (MDR)?
The MHRA’s approach shares foundational principles with the EU MDR, emphasizing safety, performance, and quality management systems, and both require robust clinical evidence and post-market surveillance. However, the EU AI Act introduces a specific high-risk classification for healthcare AI with additional obligations beyond the MDR, focusing on data governance, transparency, human oversight, and cybersecurity, which may differ in detail from MHRA’s mechanisms.
What is the significance of national regulators like the MHRA in the broader international AI regulation landscape?
Individual national regulators often act as early indicators of evolving standards, even as global harmonization remains an aspirational goal. The MHRA engages in international harmonization efforts, aligning with organizations like the WHO and FDA on principles such as Good Machine Learning Practice (GMLP), suggesting its positions often reflect broader consensus or emerging best practices that can influence larger international frameworks.
What are some key areas of focus for the EU AI Act regarding high-risk healthcare AI systems?
The EU AI Act introduces specific requirements for high-risk healthcare AI systems, focusing on data governance, transparency, human oversight, and cybersecurity. These obligations are in addition to those in the MDR and have compliance deadlines generally set for December 2027 or August 2028 depending on classification.
What challenges do global companies face in navigating the differing AI regulations in Europe?
Global companies like Tempus AI and Butterfly Network face significant operational challenges in navigating the dual compliance requirements of frameworks like the MHRA’s guidance and the EU MDR, especially with the additional obligations imposed by the EU AI Act. This requires understanding where regulatory intent aligns and where national approaches carve out unique pathways.