The year 2024 has shifted several structural issues for the healthcare sector, not on the ground of technological promise, but on that of compliance and data governance. Professionals following health news have been able to gauge the extent of European regulatory changes, often more decisive for daily practice than product announcements.
AI Act and medical devices: what the European regulation changes concretely
The regulation (EU) 2024/1689, published in the Official Journal of the European Union on July 12, 2024, classifies the majority of AI systems integrated into medical devices as high-risk systems. The nuance is technical but carries significant consequences: a diagnostic aid software or a triage algorithm is no longer solely subject to the medical device regulation. It must now meet a second set of requirements.
We observe that this dual regulatory layer requires manufacturers to formally document several areas: training data, algorithmic risk management, human oversight, cybersecurity, and post-marketing surveillance. For a medical software publisher, this means structuring a technical file that is considerably thicker than before the AI Act came into force. Professionals can find these regulatory developments and other analyses on health news on Santé 365.
The often underestimated point concerns mandatory human oversight. The AI Act requires that a healthcare professional be able to understand the system’s outputs and, if necessary, bypass them. This constraint renders certain so-called “autonomous” models that operated as black boxes obsolete.
Impact on hospital workflows
Establishments that used AI tools in production without formal traceability must review their purchasing and validation processes. Compliance is no longer verified solely at acquisition: it is maintained over time through continuous post-marketing monitoring.
In practice, hospital information system departments become co-responsible for documentation. We recommend integrating the requirements of the AI Act into the specifications from the bidding phase, rather than treating them as a late corrective measure.
European Health Data Space: governance before sharing
The regulation (EU) 2025/327 concerning the European Health Data Space (EHDS) establishes an infrastructure that far exceeds the simple exchange of patient records between practitioners. The EHDS organizes the secondary reuse of health data for research, pharmaceutical innovation, and public policy development.
Two components coexist. The primary component facilitates cross-border access for patients to their own medical data. The secondary component, less publicized, opens regulated access to anonymized or pseudonymized datasets for research organizations, regulatory agencies, or industrial players.
- Hospital care data, prescriptions, and reimbursements become accessible under strict conditions through national access organizations.
- Each member state must designate a competent authority to process secondary access requests and verify the declared purpose.
- Patients retain the right to object to secondary reuse, which requires information systems to manage opt-outs on a large scale.
For France, where the Health Data Hub already anticipated this logic, the EHDS adds a cross-border dimension and harmonizes access rules. French researchers will eventually be able to cross national cohorts with data from other member states, provided they comply with the common governance framework.
Generative AI in medicine: real adoption versus marketing discourse
The gap between laboratory demonstrations and clinical deployment remains significant. Language models applied to drafting medical reports or synthesizing scientific literature are progressing, but their integration into workflows remains marginal in France.
The main reason is not technological. It lies in the absence of a clinical validation framework suitable for generative models. A tool that produces text cannot be certified like a traditional medical device. Regulatory agencies have not yet stabilized an evaluation methodology for these systems, which hinders cautious establishments.
Use cases that are already working
Some targeted applications show tangible results. The pre-drafting of discharge letters, assisted coding of medical acts, or structured extraction of information from free-text reports are tasks where generative AI provides measurable time savings without direct clinical risk.
We observe that successful projects share a common point: they do not automate medical decision-making. They accelerate administrative tasks under the supervision of a professional. Generative AI works better as a documentary support than as a diagnostic aid.
Precision medicine and omics data: the IT bottleneck
Large-scale genomic sequencing has become financially accessible. The bottleneck now lies downstream: storing, analyzing, and returning omics data within a timeframe compatible with clinical decision-making remains an infrastructure challenge.
Hospital bioinformatics platforms often lack computing power and qualified personnel to handle the volumes generated by high-throughput sequencing. The gap between the cost of sequencing and the cost of interpretation widens each year.
- The secure storage of genomic data requires certified environments for hosting health data, which limits technical options.
- The interpretation of variants of uncertain significance requires rare skills, at the intersection of medical genetics and bioinformatics.
- The integration of omics results into the electronic patient record remains artisanal in most French establishments.
Precision medicine is no longer hindered by the cost of the test, but by the ability of information systems to absorb and exploit the results. Future investments must target infrastructure as much as research.
The European regulatory framework established in 2024 provides the rails. The challenge in the coming months for healthcare professionals and medical technology developers lies in operational compliance and in upskilling teams on topics that are no longer optional.
