In March 2026, legal commentary reported that the EU Council and Parliament had proposed delaying the full application of the EU AI Act for high-risk AI-based medical devices by 12 months, to 2 August 2028. Reuters reported on 7 May 2026 that EU lawmakers and countries had reached a provisional agreement on revised AI rules, including delayed implementation for high-risk AI systems — part of a broader effort to simplify the regulatory framework while maintaining safety requirements.
The delay does not make regulation less important. It gives serious healthcare AI companies more time to build governance, post-market monitoring, transparency, and documentation into the product itself — rather than treating compliance as a last-minute scramble.
What Changed
The EU AI Act classifies AI systems used as safety components of medical devices, or that are themselves medical devices, as high-risk. These systems face requirements including conformity assessment, risk management, quality management systems, data governance, transparency, human oversight, post-market monitoring, and detailed technical documentation.
The proposed 12-month delay for high-risk AI medical devices — from August 2027 to August 2028 — gives manufacturers additional time to prepare. But the direction of travel is clear: healthcare AI will face structured regulatory requirements across the EU, covering both pre-market assessment and post-market surveillance.
Why Delay Is Not Deregulation
Companies that interpret extra time as permission to ignore governance will find themselves unprepared when the requirements take effect — and may face market access barriers in the EU's 27 member states. The practical implications for healthcare AI companies include: classification uncertainty (determining whether a clinical AI tool falls within the high-risk category requires detailed analysis of intended use and clinical impact), evidence generation (demonstrating safety and efficacy through clinical evaluation requires data collection that takes time to accumulate), quality management systems (building QMS infrastructure that meets EU MDR and AI Act requirements is an organisational investment, not a documentation exercise), post-market surveillance (designing monitoring systems that track real-world performance after deployment), human oversight (ensuring that clinicians retain meaningful control over AI-assisted decisions), transparency (providing clear information about how the system works, what data it uses, and what its limitations are), and documentation burden (creating the technical documentation required for conformity assessment).
UK Relevance
The UK is not bound by the EU AI Act — but the direction of travel affects UK companies in multiple ways. UK companies selling into EU markets must comply with EU requirements. The MHRA is developing its own regulatory framework for AI as a medical device, and UK regulatory expectations are likely to evolve in parallel with (though not identical to) EU requirements. International clinical AI companies will increasingly build governance infrastructure that meets the most demanding regulatory requirements — making governance a baseline expectation rather than a competitive differentiator.
The MHRA already recognises that software, including AI, may be regulated as a medical device in the UK depending on intended use. UK clinical AI companies that build governance infrastructure now — clinical safety cases, post-market surveillance, transparency documentation, feedback mechanisms — will be better positioned regardless of how UK-specific regulation evolves.
iatroX Positioning
iatroX should present governance as a product feature, not as a compliance burden. Professional-facing design, source provenance, algorithmic fidelity controls, fail-safe behaviour, and feedback loops are not only safety claims — they are the infrastructure that serious clinical AI products need as regulation matures. UKCA-marked, MHRA-registered, with publicly described clinical AI standards.
Choose clinical AI that is built for verification, governance, and professional use from the start →