Artificial intelligence is reshaping healthcare at a pace that regulators, providers, and patients are still scrambling to absorb. Clinical decision-support engines, diagnostic imaging algorithms, and predictive readmission tools now routinely process some of the most sensitive data on the planet — protected health information (PHI). For AI companies operating in or expanding into France, one compliance requirement sits squarely in their path: Hébergeur de Données de Santé (HDS) certification.
HDS is not a checkbox. It is a rigorous, government-backed certification framework that dictates exactly how health data must be hosted, managed, and secured. For AI providers, HDS intersects directly with emerging AI governance standards like ISO 42001:2023, creating both a compliance challenge and a significant competitive differentiator. This guide walks you through everything you need to know — requirements, process, costs, timelines, and the smart overlap with ISO 42001 that most competitors aren't talking about.
What Is HDS Certification?
HDS (Hébergeur de Données de Santé) is a mandatory French certification framework established under Article L.1111-8 of the French Public Health Code. Any organization — French or foreign — that hosts, processes, or manages personal health data (données de santé à caractère personnel) on behalf of healthcare professionals or institutions operating in France must hold HDS certification.
The certification is issued by accredited auditing bodies under the authority of the Agence du Numérique en Santé (ANS), the French national health digital agency. It is not voluntary. Operating without it while handling French health data is a criminal offense under French law.
Citation Hook: HDS certification is a legal prerequisite — not an optional credential — for any entity hosting personal health data on behalf of French healthcare actors, regardless of whether that entity is physically located in France.
There are two scopes of HDS certification:
| Scope | Who It Applies To | Key Focus |
|---|---|---|
| Infrastructure Hosting | Data centers, cloud IaaS/PaaS providers | Physical and virtual infrastructure security |
| Software Application Hosting | SaaS and application-layer providers | Software lifecycle, access controls, data management |
Most AI providers delivering healthcare SaaS or API-based services fall under Scope 2 (Software Application Hosting), though companies operating their own cloud infrastructure may need both.
Why HDS Certification Matters Specifically for AI Providers
AI providers occupy a uniquely exposed position in the health data ecosystem. Unlike traditional EHR vendors that store and retrieve structured records, AI systems analyze, infer, and derive new information from PHI. That creates a broader attack surface and a higher liability profile.
Consider the following:
- The French health data market is substantial. France's national health insurance system (Assurance Maladie) covers over 67 million people, and digital health spending in France exceeded €2 billion in 2023, according to the French Ministry of Health's digital transformation roadmap.
- AI health applications are growing fast. The global AI in healthcare market was valued at approximately $20.9 billion in 2024 and is projected to reach $148.4 billion by 2029, according to MarketsandMarkets research — representing a CAGR of 48.1%.
- Enforcement is real. France's data protection authority, the CNIL, issued €101 million in GDPR fines in 2023 alone, and HDS violations can trigger parallel criminal proceedings under the Public Health Code.
- Trust signals win contracts. A 2024 survey by Statista found that 78% of healthcare procurement officers in Europe cited third-party security certification as a primary vendor selection criterion.
For an AI provider, HDS certification is the single most credible signal you can send to French hospital groups, insurers, and public health agencies that you take PHI protection seriously.
HDS Certification Requirements: What Auditors Actually Look For
HDS certification is built on a base of ISO/IEC 27001:2022 (information security management) combined with additional health-specific requirements defined in the ANS reference framework. For AI providers, the most demanding requirements cluster around six domains:
1. Governance and Contractual Obligations
You must have formal, written contracts (hosting agreements) with each healthcare client that explicitly define data processing responsibilities, security obligations, and breach notification procedures. For AI providers, this means your standard SaaS agreements almost certainly need legal revision before an audit.
2. Physical and Logical Access Control
All environments processing health data must implement strict logical segregation, multi-factor authentication, and role-based access controls. AI training pipelines that use production health data — even in anonymized form — require documented controls and justification.
3. Data Lifecycle Management
HDS auditors scrutinize the entire data lifecycle: ingestion, processing, storage, archival, and secure deletion. For AI systems, this includes training data, model weights derived from health data, inference logs, and API transaction records. Many AI providers are surprised to learn that model artifacts may be considered derivative health data under French interpretations.
4. Business Continuity and Disaster Recovery
You must demonstrate tested BCP/DR plans with defined Recovery Time Objectives (RTOs) and Recovery Point Objectives (RPOs) appropriate for health data. Regulators expect RTOs of 4 hours or less for critical systems.
5. Incident Management and Breach Notification
Breach notification to ANS must occur within 72 hours of discovery — consistent with GDPR Article 33 requirements. AI providers must instrument their systems to detect anomalous data access patterns, which is often more complex in ML inference environments than in traditional databases.
6. Subcontractor Management
If your AI platform relies on third-party cloud infrastructure, API services, or data annotation providers, every subcontractor touching health data must themselves be HDS certified or operate under equivalent controls you can audit and document. This is a common failure point for AI startups using mixed cloud architectures.
The HDS Certification Process: Step by Step
Here is the realistic, practitioner-level roadmap for an AI provider pursuing HDS certification:
Step 1: Scope Definition (Weeks 1–3)
Map every system, process, and third party that touches French health data. Define your certification boundary clearly. Overclaiming scope inflates audit cost; underclaiming creates certification gaps that surface later.
Step 2: Gap Assessment Against ISO 27001 + HDS Add-ons (Weeks 3–8)
If you don't already hold ISO 27001 certification, this gap assessment will reveal the distance you need to travel. Most AI providers without an existing ISMS find 60–80 controls in partial or non-conformance. Document every finding with a remediation owner and target date.
Step 3: ISMS Implementation and Remediation (Weeks 8–24)
Build or mature your Information Security Management System. For AI providers, this phase must explicitly address: - AI-specific risks (model poisoning, adversarial inputs, data leakage through model outputs) - Data minimization for training datasets - Logging and auditability of AI inference decisions affecting patient care
Step 4: Internal Audit (Weeks 24–28)
Conduct a full internal audit of your ISMS and HDS-specific controls. Use the ANS reference framework as your audit checklist. Address all non-conformances before bringing in the external auditor.
Step 5: Stage 1 Audit — Document Review (Weeks 28–32)
An ANS-accredited certification body reviews your documentation. Expect scrutiny of your risk assessment methodology, Statement of Applicability (SoA), and health-data-specific procedures.
Step 6: Stage 2 Audit — On-Site/Remote Assessment (Weeks 32–38)
Auditors verify implementation through interviews, system demonstrations, and evidence sampling. For AI providers, be prepared to walk auditors through your ML pipeline, data ingestion process, and how PHI is isolated in training vs. inference environments.
Step 7: Certification Decision and Issuance (Weeks 38–42)
Assuming no major non-conformances, your certification is issued for a three-year period with annual surveillance audits.
Realistic total timeline: 9–12 months for a well-resourced AI provider starting from scratch.
HDS and ISO 42001: The Dual-Certification Advantage
This is where I see most competitors' content fall short — they treat HDS and ISO 42001 as entirely separate tracks. They're not.
ISO 42001:2023 is the international standard for AI Management Systems (AIMS). It establishes requirements for responsible AI development, deployment, and governance — covering risk management (clause 6.1), impact assessment (clause 6.1.4), and AI-specific controls across the AI system lifecycle (Annex A).
For healthcare AI providers, pursuing HDS and ISO 42001 in an integrated program delivers compounding benefits:
| Requirement Area | HDS Requirement | ISO 42001:2023 Equivalent | Overlap Opportunity |
|---|---|---|---|
| Risk Management | Health data risk assessment | Clause 6.1.2 — AI risk identification | Unified risk register covering both |
| Incident Response | 72-hour breach notification | Clause 8.4 — AI incident management | Single incident management procedure |
| Subcontractor Control | HDS-certified suppliers | Clause 8.6 — AI supply chain | Combined supplier assessment program |
| Documentation | HDS hosting agreement | Clause 7.5 — documented information | Shared document control system |
| Audit & Review | Annual surveillance audits | Clause 9.2 — internal audit | Coordinated audit calendar |
| Data Lifecycle | Health data retention/deletion | Annex A, Control A.6.2 — data for AI | Integrated data governance policy |
Citation Hook: AI providers that pursue HDS certification alongside ISO 42001:2023 can reduce total compliance implementation effort by an estimated 30–40% through shared control frameworks, unified risk registers, and coordinated audit programs.
I've guided organizations through exactly this dual-certification approach at Certify Consulting, and the efficiency gains are real and measurable. The key is sequencing: establish your ISO 27001 foundation first, layer in HDS health-specific requirements, then extend your AIMS governance upward using ISO 42001's clause structure.
For a deeper look at how ISO 42001 governance requirements map to your AI platform, see our ISO 42001 implementation guide for healthcare AI.
Common Failure Points for AI Providers Pursuing HDS
Based on my experience with 200+ compliance engagements across regulated industries, these are the most common places AI providers stumble in the HDS process:
1. Training Data Provenance Auditors will ask where your training data came from, how it was licensed, whether it contained real PHI, and how it was de-identified. Many AI startups cannot answer these questions with documentation. If your model was trained on a public dataset derived from clinical records, you need a paper trail.
2. Model Artifact Classification Does a model trained on patient data constitute "health data"? French regulators are increasingly taking the position that it might. Get ahead of this by classifying model artifacts in your data inventory and applying appropriate access controls.
3. Inference Log Retention Every API call that processes patient data is, technically, a health data processing event. Your logging architecture must capture these events, retain them appropriately, and make them available for audit without exposing patient data to unauthorized parties.
4. Multi-Tenant Architecture Risks SaaS AI platforms serving multiple hospital clients must demonstrate that PHI from one client cannot be accessed by — or inadvertently exposed to — another. Logical isolation must be documented and tested, not assumed.
5. Subcontractor Certification Gaps Using a non-HDS-certified LLM API provider to process French patient queries is a certification-breaking finding. Audit your entire dependency stack before your Stage 1 audit.
Cost and ROI Considerations
HDS certification is an investment. Here's a realistic cost picture for a mid-sized AI provider:
| Cost Component | Estimated Range (EUR) |
|---|---|
| Gap assessment and consulting | €15,000 – €40,000 |
| ISMS implementation (if building from scratch) | €30,000 – €80,000 |
| External audit fees (certification body) | €20,000 – €45,000 |
| Legal review of hosting agreements | €5,000 – €15,000 |
| Annual surveillance audit | €8,000 – €15,000/year |
| Total Year 1 Investment | €78,000 – €195,000 |
These numbers are significant, but context matters. A single contract with a major French hospital group or insurer typically represents €500,000–€2M+ in annual recurring revenue. HDS certification is frequently the gating requirement for those contracts. The ROI calculation is not complicated.
Citation Hook: For AI providers targeting the French healthcare market, HDS certification is not a cost of compliance — it is the cost of market entry, and organizations that delay certification consistently lose contracts to certified competitors.
What Schellman and Others Get Right — and Where to Go Deeper
Schellman's analysis of HDS benefits for AI providers (published on schellman.com) correctly identifies the foundational value of HDS certification: building trust, demonstrating commitment to PHI protection, and enabling market access in France. That's a solid foundation.
What deeper practitioner experience adds is the operational specificity that matters most to AI engineering and compliance teams:
- The intersection of model artifact governance with HDS data lifecycle requirements
- The dual-certification efficiency of pairing HDS with ISO 42001:2023
- The inference log architecture decisions that determine audit readiness
- The subcontractor dependency risks unique to AI platforms
These are not theoretical concerns. They are the findings that surface in real audits, and preparing for them is the difference between a first-time pass and a remediation cycle that costs months and six figures.
Is HDS Certification Right for Your AI Platform Now?
Ask yourself these questions:
- Do you have any current or prospective French healthcare clients — including public hospitals, private clinics, insurers, or medical software vendors?
- Does your AI platform process, analyze, or store data that could be traced back to individual patients?
- Are you building partnerships with French digital health platforms that are themselves HDS certified?
- Are you pursuing EU healthcare AI contracts where French regulatory frameworks carry influence?
If you answered yes to any of these, the question is not whether to pursue HDS certification — it's when and how efficiently.
Our team at Certify Consulting has helped organizations achieve ISO 27001, ISO 42001, and health-specific certifications with a 100% first-time audit pass rate across 200+ engagements. We understand how to sequence your compliance investments for maximum efficiency and minimum audit risk.
To explore how HDS and ISO 42001 certification can be integrated into a single, streamlined program for your AI platform, visit iso42001consultant.com or contact us directly at certify.consulting.
Conclusion
HDS certification is a non-negotiable requirement for AI providers handling French personal health data — and it is increasingly a competitive advantage in the broader European health AI market. The certification process is demanding, but it is navigable with the right expertise and the right sequencing.
The most forward-thinking AI providers are not treating HDS as a standalone compliance burden. They are integrating it with ISO 42001:2023 AI management system requirements to build a unified governance posture that satisfies auditors, reassures hospital procurement officers, and future-proofs their platforms against emerging AI-specific health data regulations.
That integrated approach is exactly what we help our clients build at Certify Consulting — efficiently, defensibly, and with a track record that speaks for itself.
Last updated: 2026-04-07
Jared Clark
Principal Consultant, Certify Consulting
Jared Clark is the founder of Certify Consulting, helping organizations achieve and maintain compliance with international standards and regulatory requirements.