The genome has gone from museum exhibit to moving conveyor belt. Sequencers now decode three billion base pairs at a unit cost that would have looked absurd a decade ago, and the next layer of machinery, autonomous AI agents that interpret variants and trigger clinical actions, is already brushing against the walls of European hospitals. The question is whether Europe will let that machinery in, slow it at the door, or watch it grow up somewhere else.
A market that compounded in silence
Global sequencing throughput has compounded at double digits for ten consecutive years. Illumina alone has shipped enough instruments to enable, in figures collated by McKinsey, somewhere north of 60 million human genomes sequenced worldwide by the close of 2025. The European Investment Bank has projected that the bloc’s bioeconomy will absorb around 100 billion euros of capex into life sciences by 2030, with genomic infrastructure taking a meaningful slice. Underneath those numbers a quieter shift has taken place. Sequencing is no longer the bottleneck. Interpretation is. A single whole genome generates around 100 gigabases of raw signal, and a clinical-grade variant call set typically holds three to five million single nucleotide variants per patient. No human team can read that map alone. AI is the only plausible cartographer, and within AI, the agent is the only plausible field surveyor.
What an autonomous agent would actually do
Strip away the hype and an autonomous genomic agent looks mundane. It would ingest a patient’s variant file, cross reference it against ClinVar, gnomAD, and proprietary curated panels, prioritize variants of uncertain significance, draft a structured interpretation, propose a confirmatory assay, schedule the follow up consultation, and write the first draft of the clinical note. In oncology it would suggest a targeted therapy chosen from a constellation of NCCN, ESMO, and label-restricted options. In rare disease it would surface candidate diagnoses with their likelihood ratios. None of this is science fiction. Tempus, Foundation Medicine, and a growing pack of European startups already run pipelines where the clinician’s role is increasingly editorial rather than authorial.
The shift, then, is not whether machines participate in the diagnosis. It is whether they sign it.
Europe’s regulatory thicket
This is where the rails slow. The European Health Data Space, agreed in early 2024 and now phasing in across member states, finally gives clinicians and researchers a continent-wide framework for secondary use of genomic data. EHDS will, according to European Commission impact assessments, save the bloc somewhere around 11 billion euros over ten years through better data flow. That is good news. The less good news is that the AI Act classifies any AI system used to triage or diagnose patients as high risk under Annex III, requiring conformity assessment, human oversight, post market monitoring, and traceable training data. Pair that with the In Vitro Diagnostic Regulation, which has already pushed thousands of legacy assays into a notified body bottleneck, and a clinical AI must clear three distinct regulatory gates before a single patient sees a result.
Then comes the harder question, that of agency itself. GDPR Article 22 still bars solely automated decisions with significant effects in most clinical contexts. The proposed AI Liability Directive, paired with the revised Product Liability Directive, shifts the burden of proof toward providers when a high risk system causes harm. Translate that into a clinician’s daily life and the message becomes blunt, an autonomous agent that drafts a recommendation is welcome, an autonomous agent that issues one is a legal hazard.
How the United States and China play different games
The United States is running a faster, messier experiment. The FDA’s Digital Health Center of Excellence has issued more than 950 AI and machine learning enabled device authorizations to date, including a long tail of clinical decision support tools. The 21st Century Cures Act removed most clinical decision software from device classification when a human stays in the loop, which lets a Tempus-style pipeline iterate at venture speed. Reimbursement is still a friction, yet the regulatory floor is low and the litigation system handles edge cases case by case.
China is running yet another game. The Cyberspace Administration’s generative AI rules apply on top of an NMPA framework that has issued explicit guidance for AI as a medical device since 2022 and has approved more than 80 AI assisted diagnostic products by mid 2026, in figures reported by the Stanford AI Index. Genomics sits inside the larger industrial blueprint, with BGI scaling population sequencing programs few European peers can match in raw volume. The state acts as both regulator and customer, and the speed of deployment reflects that arrangement.
The contrast is not subtle. Europe demands proof before scale. The United States allows scale and litigates exceptions. China commands scale and writes the rulebook around it.
Spain as a stress test for the European model
Spain offers a useful lens onto the European setup because it has both serious research depth and a fragmented payer landscape. The Instituto de Salud Carlos III coordinates IMPaCT, a national precision medicine program that, in figures attributed to the program itself, aims to sequence around 200,000 genomes by the end of the decade. The IDIBAPS biobank in Barcelona and the network of regional biobanks under the ISCIII umbrella give clinicians a base of curated samples few European countries can rival. Vall d’Hebron and Hospital Clínic have piloted AI assisted variant interpretation. Grifols, headquartered in Sant Cugat, has built a global plasma and diagnostics franchise that hints at what a Spanish RegTech for health champion could look like with the right national policy alignment.
The catch is governance. AEMPS must coordinate with seventeen autonomous communities. Each region runs its own electronic health record, its own ethics committee, its own procurement cycle. An autonomous genomic agent that works in Catalonia must clear another door in Madrid and a different one in Andalusia. That is not unique to Spain. It is the European condition reproduced at country scale.
The European thesis
The lazy reading of all this is another chapter of European decline, the slow continent watching faster ones eat the future. That reading does not survive contact with the data. Europe’s caution around autonomous medical agents is not a bug, it is a hedge. A genomic agent that confidently miscalls a BRCA variant in 2027 would produce harm at industrial scale, the way a faulty brake pad recalled across a million vehicles compounds risk far beyond any single defect. The bloc that lays the legal and procedural rails first will be the one that exports those rails later, much as GDPR became a global template even where it was resented.
The real risk is not that Europe is slow. The real risk is that Europe is slow without a corresponding industrial bet. Asking clinicians to use only certified AI agents while ceding the supply of those agents to American and Chinese vendors would replicate the cloud dependency story in a domain where data sovereignty matters more, not less. EHDS gives Europe a unique substrate. The AI Act gives it a tilted, enforceable bar. What is still missing is a credible industrial vehicle, EIB credit lines for medical AI scaleups, mandatory preference for EU-certified systems in public procurement, a clear pathway for autonomous decision support to graduate from advisory to assisted and, eventually, to autonomous under sectoral oversight.
The autonomous agent will reach the clinic this decade. The question Europe still has to answer is whether it will be built in Munich, Barcelona, and Stockholm, or shipped from Boston and Shenzhen with a CE mark stapled to its side.
Sources
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), Annex III. Official Journal of the European Union.
Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data (General Data Protection Regulation), Art. 22. Official Journal of the European Union.
Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products (recast) (Product Liability Directive). Official Journal of the European Union.
European Commission. (2022, September 28). Proposal for a Directive of the European Parliament and of the Council on adapting non-contractual civil liability rules to artificial intelligence (AI Liability Directive) (COM/2022/496 final). EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52022PC0496
European Commission. (2022). Commission Staff Working Document: Executive Summary of the Impact Assessment Report Accompanying the document Proposal for a Regulation of the European Parliament and of the Council on the European Health Data Space (SWD/2022/132 final). EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52022SC0132
U.S. Food and Drug Administration. (2026, January). Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff. FDA. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software
National Medical Products Administration (NMPA). (2022, March 9). Guideline for Artificial Intelligence in Medical Devices Registration (Review Guideline No. 8). China Medical Device. https://chinameddevice.com/guideline-on-artificial-intelligence-medical-devices/
Stanford University, Institute for Human-Centered Artificial Intelligence (HAI). (2026). Artificial Intelligence Index Report 2026 (Chapter 6: Medical Applications). https://hai.stanford.edu/ai-index
