The Member State Coordination Group on Health Technology Assessment (HTACG) has adopted general principles for artificial intelligence used in evidence synthesis submitted for EU Joint Clinical Assessments (JCAs). The principles apply to all machine-learning models, including generative AI, used in literature searching, study screening, data extraction, risk-of-bias assessment, analysis or reporting.
JCAs provide an EU-level scientific assessment of the relative clinical effects of medicines and medical devices, supporting later national HTA processes. The initial scope has covered new cancer medicines and advanced therapy medicinal products since 12 January 2025.
HTACG recognises that AI may improve the efficiency of evidence retrieval, analysis and reporting. Developers nevertheless remain accountable for whether and how it is used, the validity of its outputs and its effect on the evidence synthesis. Responsibility also covers copyright, data protection and applicable requirements under the EU Artificial Intelligence Act.
Human oversight must continue throughout each AI-assisted stage. An identifiable person must remain responsible for quality and accuracy, while existing HTACG methodological requirements continue to apply. Developers may therefore need to show how AI-supported processes were checked for completeness, reproducibility and error.
All AI-assisted stages must be identified in the dossier. Disclosure extends to outputs or judgements that were generated or informed by AI, potentially covering screening, extraction, classification and methodological support as well as generated text.
Dossiers should report the tool name, version and date, developer and purpose. Adaptations to commercial tools should preferably also be described. All prompts must be recorded and made available if requested during the JCA.
Prompt records are likely to become part of the dossier audit trail, requiring version control, document retention and quality-assurance procedures. Developers will also remain accountable where consultancies, evidence-review suppliers or medical writers use AI on their behalf.
The principles leave detailed implementation questions open, including disclosure format, treatment of model updates, reporting of proprietary tools and task-specific validation standards. HTACG may issue further guidance as experience develops.
Source: European Commission
Link: General Principles on the use of Artificial Intelligence in the preparation of the JCA dossier
Date: 16 July 2026
