The cost of reading a human genome has fallen so far that the accountant, not the geneticist, is now the bottleneck. Illumina’s NovaSeq X platform, deployed in fourteen European reference labs by early 2025, has pushed whole genome sequencing below 200 euros per sample, and Ultima Genomics is testing sub-100-dollar runs. What was once a moonshot has quietly become a commodity. The interesting question in 2026 is no longer how to read the book. It is who gets to act on what is written inside, at what speed, and with which signature at the bottom of the order.
A continent reading itself
Europe has decided to read itself with unusual discipline. The 1+ Million Genomes initiative, first declared in 2018 by twenty-seven signatory countries, adopted a fresh roadmap in November 2023 and expects fifteen national infrastructures for genomic data access to be operational by 2026, according to the European Commission’s Digital Strategy service. A 20-million-euro grant now funds the Genome of Europe project across twenty-six Member States, targeting a reference genome built from around 100,000 representative citizens. The European Health Data Space, in force since March 2025, will begin to interconnect eighteen national repositories covering more than two million sequenced individuals from 2026 and, per Kennedys’ summary of the regulation, only opens the tap for secondary use of genomic data in earnest from March 2031.
That last date matters. The paperwork will trail the pipette by half a decade. In the meantime, the market races on. Fortune Business Insights placed Europe’s genomics segment at roughly 9.45 billion dollars in 2025 and 10.6 billion in 2026, while Next Generation Sequencing revenues in Europe are forecast to compound at just over twenty-one percent through 2034. If a market is a river, this one has been rerouted; the banks, drawn by regulators in Brussels and Strasbourg, are still being reinforced upstream.
Agents that can read a genome and cannot yet order the biopsy
Enter the autonomous agents. Variant Bio launched Inference in January 2026, marketed as the first agentic AI platform for genomic drug discovery, with in-loop agents that plan queries, run pipelines, and interpret results without a human in every step. Academic teams have built systems, GenomixIQ among them, that classify a variant in eight seconds against curated evidence bases. In a well-designed clinical workflow, such an agent could triage a suspected rare disease case overnight, pull the relevant literature, propose the confirmatory panel, and hand the report to a human geneticist by breakfast.
In Europe, the geneticist still has to sign every meaningful outcome. Article 22 of the GDPR, reinforced by the Court of Justice of the European Union’s SCHUFA ruling in December 2023, treats fully automated decisions with legal or similarly significant effects as presumptively prohibited outside narrow gateways. Under the Medical Device Regulation, software that interprets variants is a medical device, and under the AI Act any such device or safety component drops directly into Annex III as high risk. The Digital Omnibus adopted by the Council on 29 June 2026 pushed the standalone high-risk deadline to 2 December 2027 and the medical device deadline to 2 August 2028, but pushed no principle. Autonomy without a nameable, insured human signatory remains, in effect, unwelcome at the bedside.
The proposed AI Liability Directive, which reallocates the burden of proof in favor of injured patients, and the recast Product Liability Directive, which brings software squarely inside the strict liability regime, sharpen the point. An agent that recommends a targeted therapy for a variant of uncertain significance is a fine junior colleague. It is a poor named defendant.
Two other continents, two other tempos
The United States is running the same experiment with a different tolerance for uncertainty. The FDA, in guidance updated in January 2026, has tightened the perimeter around clinical decision support that behaves autonomously or is heavily driven by generative models, while continuing to clear specific SaMD products one at a time. The result is a patchwork with a bias toward speed, in which private hospital systems can pilot agentic genomic workflows under IRB cover long before a central rulebook catches up. Innolitics documented the first FDA-cleared AI agent in a medical device context earlier this year, a signal that the American regulator can move when a use case is bounded.
China has taken the opposite route, walling the garden and then irrigating it heavily. Precision medicine is a headline commitment in the current five-year plan, backed by more than nine billion dollars of state research spending, with BGI operating the largest single sequencing footprint in the world. On 2 April 2026, the National Medical Products Administration published its Implementation Opinions on AI plus Drug Regulation, with milestones stretching to 2035, while the Human Genetic Resources Administration of China continues to gate every foreign use of Chinese genomic samples. The playbook is unmistakable. Run the pilots at scale inside the wall, decide later what leaves it.
Europe, held to its own procedural code, risks a familiar outcome. Clinical trials involving agentic genomics may migrate to Boston or Shenzhen, while European hospitals and biotechs pay the tax of running the same workflows with more humans, more forms, and more delay. That is the productivity gap the Draghi report warned about, only translated from euros into base pairs.
Spain as a stress test
Spain is a useful microcosm. The IMPaCT infrastructure, coordinated by the Instituto de Salud Carlos III, is building a 200,000-person cohort tracked over twenty years through fifty primary care centers, of which thirty are already active with close to seven thousand participants recruited, according to ISCIII figures. In July 2026, the Council of Ministers approved a fresh 50-million-euro envelope for the 2026 to 2030 IMPaCT window. The program’s genomic axis, IMPaCT Genómica, applies complex genome analysis to undiagnosed rare disease cases, and its data axis, IMPaCT Data, does what the acronym suggests.
An agentic layer would be the obvious next step. A rare disease agent could sit on top of IMPaCT Data, ingest a new case from Vall d’Hebron or a regional hospital, run a variant classifier, cross-reference IMPaCT Cohorte, and propose a diagnostic route within minutes. Yet the same agent would run headlong into three constraints: the AEMPS as a medical device gatekeeper, the AI Act as a horizontal risk regime, and the Spanish transposition of the AI Liability Directive still being drafted. The answer, in practice, will be a human in the loop with an electronic signature. The question is whether the loop can be closed in hours rather than weeks.
The European wager
Europe is betting that a continent can move fast on infrastructure and slow on autonomy without losing the race. The wager is not obviously wrong. Trust in a genomic result is a currency of its own, and Europe still prints it at higher denominations than most jurisdictions. The risk is that the world learns to trust American and Chinese agents first, because they were the ones allowed to make mistakes in public. If the AI Liability Directive is passed in a workable form, if the EHDS delivers its 2027 milestones on time, and if national regulators like AEMPS, ANSM, BfArM, and MHRA converge on a shared playbook for agentic medical software, the delay will look like prudence. If not, Europe will have built the largest genomic library ever assembled and hired someone else to read it aloud.
Sources
- 1+ Million Genomes roadmap (European Commission)
- EHDS implementation timeline (Kennedys Law)
- EU AI Act and Medical Devices deadlines (EducoLifeSciences)
- IMPaCT 2026 program (ISCIII)
- NMPA AI + Drug Regulation Opinions (Cisema)
- Variant Bio Inference agentic platform
- Europe personalized medicine market forecast
