AI in healthcare, within medical secrecy.
AI consulting and training for physicians, nurses, pharmacists, medical biologists, healthcare and social care institutions. Reports, anonymisation, monitoring developments in treatments. HDS-certified hosting or local processing for health data.
Colombani.ai helps healthcare professionals and institutions use AI to structure medical reports, anonymise patient files and monitor developments in treatments. Health data is processed locally or by an HDS-certified host. The framework combines reinforced GDPR requirements, professional secrecy (art. L.1110-4 CSP) and the AI Act, which classifies only certain healthcare uses as high-risk.
Use cases
Clinical reports and summaries
First version of consultation, surgical, lab reports from voice or text notes. Processing runs locally, without data leaving your machines.
Patient record anonymisation
Pseudonymisation for multidisciplinary review meetings, internal training, clinical research. Preserves relevant medical content, removes direct and indirect identifiers.
Therapeutic and bibliographic monitoring
Weekly synthesis of PubMed, Cochrane, HAS publications on your specialty. Sourced, dated, exportable.
Plain-language patient communication
Reformulation of complex reports into accessible explanations for the patient or family. Medical content stays under your control; AI only adapts the wording.
Ethics and regulatory framework
Medical secrecy and HDS hosting
A provider that stores health data on behalf of an institution or practitioner must be HDS-certified (art. L.1111-8 CSP); processing on your own machines is the other route. No identifying data goes to a non-certified online AI service without robust pseudonymisation. Colombani.ai builds this requirement into the architecture from the design stage.
AI Act: high-risk classified uses
The EU AI Act classifies certain healthcare uses as high-risk (diagnostic aid, triage, medical devices). Enhanced obligations: documentation, risk management, human oversight. Our training covers the obligations currently in force.
What we don't do
- Medical diagnosis: reserved for licensed healthcare professionals. AI never establishes a diagnosis; it assists the clinician.
- Drug prescription: reserved for licensed physicians, who remain the sole decision-makers.
- Medical device qualification (CE, MDR, IVDR): this falls to notified bodies and the ANSM.
- HDS hosting certification: issued by accredited certifiers, not by Colombani.ai.
- Replacing the institution DPO: Colombani.ai collaborates with your DPO, never substitutes.
How do we work with you?
Consulting: AI maturity audit
We review your actual uses, data and regulatory framework, then deliver a costed roadmap. One-week Flash Diagnostic at €3,000 excl. VAT.
Training: AI for regulated professions
1 to 2 days. Healthcare case studies (reports, anonymisation, AI Act).
Consulting: sovereign architecture
Local AI or HDS-certified cloud to process patient data within the framework of the French public health code.
Consulting: AI Act compliance
Risk classification of your AI uses and compliance roadmap.
Frequently asked questions
Can I dictate my reports to ChatGPT's voice mode? +
No, not with identifying patient data. ChatGPT is not an HDS host. Solutions: (1) local dictation with Whisper + a local LLM (Ollama), (2) HDS-certified medical solution (Doctolib, Maincare), (3) prior anonymisation.
Does my DPO need to be involved? +
Yes. The processing must be recorded in the register, and the CNIL lists processing by healthcare and medico-social institutions for the care of individuals among those requiring a DPIA: an AI tool connected to patient or resident records falls within it. Our training includes a DPO module, and Colombani.ai works with your DPO on consulting engagements.
Is an HDS-certified host required to use AI on health data? +
Yes, as soon as a provider stores that data on your behalf: article L1111-8 of the French public health code then requires an HDS-certified host, including for data collected in social and medico-social care. Processing on your own servers or workstations is not hosting on behalf of a third party; prior pseudonymisation also limits the risks in the event of a leak.
AI in social and medico-social care services: where to start? +
With a review: what teams already use, the data on the people supported that they handle and the applicable framework (professional secrecy, certified hosting, impact assessment). Colombani.ai's AI maturity audit does this in one week in its Flash version, then identifies the uses to launch, to frame or to rule out. Support functions lend themselves to first trials; any assistance with a decision concerning a person supported requires specific scoping and the DPO's opinion.
What does the French National Authority for Health recommend on generative AI? +
In its “first keys for using generative AI in healthcare” (30 October 2025, updated 15 April 2026), aimed at professionals in the health, social and medico-social sectors, the HAS proposes the A.V.E.C. approach: learn, verify, estimate, communicate. The professional remains responsible for the result; this is also the principle of Colombani.ai's training.
Got a project in mind?
Describe your situation. The first 30-minute call is free. Get a frank response within one business day.