AI in JCA dossiers: the EU's new rules, explained
On 15 July 2026, the EU's HTA Coordination Group adopted its first general principles for using artificial intelligence in Joint Clinical Assessment dossiers. If your team uses AI anywhere in evidence synthesis, these principles now describe what assessors expect to see documented. Here is what they say, and how Regulaido was built to satisfy them without extra work.
A 60-second refresher: what a JCA is
Since 12 January 2025, the EU Health Technology Assessment Regulation (Regulation (EU) 2021/2282) requires a single, EU-level Joint Clinical Assessment for certain new medicines, running in parallel with EMA marketing authorisation. The JCA analyses relative clinical effectiveness against the comparators that Member States care about; pricing and reimbursement decisions remain national.
The scope is expanding in waves:
JCA scope expansion
The first JCA report, for tovorafenib (Ojemda), was published in June 2026, with more following weeks later.
For health technology developers (HTDs), a JCA dossier is a large evidence-synthesis exercise on a tight clock. AI is increasingly part of how these dossiers get built, and the HTACG has now set expectations for its use.
What the principles say
The document, General principles on the use of Artificial Intelligence in the preparation of dossiers for Joint Clinical Assessments, runs to three pages and stays deliberately high-level. It defines AI broadly, covering all machine-learning models, including generative ones, regardless of architecture. Its expectations fall into four areas:
1. Accountability stays with you
The HTD is accountable for the content, methods and findings of the dossier, including every decision about whether and how AI is used, its impact on the evidence synthesis, and the validity of any AI output. AI use must also comply with copyright, data protection (the dossier will be published), and the EU AI Act where applicable. "The AI did it" will not be a defence.
2. Human oversight, always
No step in dossier preparation or analysis may be fully automated, and a human must be ultimately responsible for the quality and accuracy of every AI-assisted step. Regulators and HTA bodies have converged on this position: the EMA's reflection paper on AI in the medicinal product lifecycle (which the HTACG cites for its AI definition) and NICE's position statement on AI in evidence generation both treat AI as support for human judgement rather than a substitute.
3. Transparency in the dossier
If AI is used, the dossier must clearly document all AI-assisted steps. The principles name information retrieval (including study screening), data extraction, risk-of-bias assessment, analysis, and reporting as examples. Any output or judgement that is AI-generated or AI-informed must be transparently identified. AI used purely for spelling and grammar is exempt from disclosure.
4. Reporting, down to your prompts
At minimum, the dossier must report the name, version and date of the AI tool used, its developer, and the specific purpose(s) it was used for. Describing adaptations of commercial tools is called out as good practice. On prompts, the wording is explicit:
"…it is expected that all prompts used in AI tools are recorded and made available if requested during the joint clinical assessment process."
HTACG general principles, July 2026
If your team's AI workflow runs through a general-purpose chatbot and ad-hoc copy-paste, reconstructing that record after the fact ranges from painful to impossible.
What this means in practice
Read as a compliance checklist, the principles mean an HTD using AI in dossier preparation should be able to answer these questions at submission, and potentially years later:
This is standard audit-trail thinking, but it only works if the record is captured at the moment of use. A tool that doesn't record it structurally leaves the whole burden on your team's discipline.
How Regulaido handles this
We built Regulaido for regulatory and HTA work, and the auditable record is designed into the system rather than left to user habit:
- Every prompt is recorded. All user prompts, AI responses, and the tools the AI invoked are persisted with timestamps, including tasks delegated to specialised sub-agents.
- Tool version and models, per interaction. Each conversation turn is stamped with the Regulaido version and the specific AI models used, so "name, version and date of the AI tool" is answerable for any point in your project's history.
- One-click disclosure reporting. Regulaido generates an AI Use Disclosure report for a project covering tool and developer identification, versions and models used, the covered period, every prompt, the AI-assisted steps taken, and sub-agent delegations, formatted as a document you can annex to your dossier workflow or hand to an assessor on request.
- Human oversight by design. Every AI contribution flows through a human-driven conversation, document edits are explicit and reviewable, and a full audit log covers the surrounding activity.
The HTACG says further guidance will follow, and we expect the requirements to become more specific over time. Transparency requirements are only a burden if your tooling wasn't built for them.
If you're preparing for a JCA and want AI assistance that produces its own compliance record, get in touch and we'll show you the disclosure report from a live project.
