A new European regulatory benchmark
The Spanish Data Protection Agency (AEPD) and the Hellenic Data Protection Authority (HDPA) have issued landmark joint guidance defining the technical and organizational governance required for Secure Processing Environments (Entornos de Tratamiento Seguro or ETS, also known internationally as Trusted Research Environments). The guidance unifies criteria across the General Data Protection Regulation (GDPR), the Data Governance Act (DGA), and the European Health Data Space (EHDS). Developed in collaboration with Eurostat, Spain's Dirección General del Dato, the Italian Garante, and the UK Information Commissioner's Office (ICO), this framework provides operational certainty for organizations training advanced machine learning models on high-value sensitive datasets.
The initiative addresses an imperative technical challenge: enabling sophisticated algorithmic research without compromising fundamental rights or citizen privacy. By precisely defining what constitutes a secure enclave, regulatory authorities offer public bodies, healthcare consortiums, and technology enterprises an auditable blueprint to collaborate on complex analytics without triggering unlawful data transfers.
The paradigm shift: compute-to-data
At the core of the framework lies the departure from traditional file exports and imperfect anonymization downloads. The guidance formalizes the 'compute-to-data' paradigm: instead of disseminating datasets to external researchers or algorithms, analytic code and training pipelines are brought directly into a shielded, monitored, and highly configured environment. Direct extraction or downloading of microdata, clinical records, or sensitive personal records is strictly prohibited across the defined trust tiers.
Under this architecture, the environment provider maintains total custody over physical and logical infrastructure. Data scientists and AI operators interact solely through controlled gateways where compute sessions are fully indexed and logged. Upon completion, researchers do not export underlying datasets, but solely reviewed mathematical model weights or aggregated statistical metrics.
Maturity of required privacy technologies
To enforce environmental isolation, the guide mandates the practical implementation of Privacy-Enhancing Technologies (PETs). Deploying federated learning architectures, formal differential privacy budgets, and confidential computing hardware allows organisations to extract deep algorithmic utility from proprietary datasets while preserving strict de-identification guarantees.
From the perspective of basebcn, adopting these methodologies is not a friction point or compute penalty, but the architectural foundation required to clear regulatory audits and commercialize enterprise AI solutions in highly regulated markets.
Architecture and defenses against model inversion
One of the most consequential mandates in the joint framework is the enforcement of dedicated output checking mechanisms tailored for machine learning. Training deep neural networks on clinical histories or biometric datasets introduces the risk of unintended memorization, leaving models vulnerable to model inversion or membership inference attacks where malicious actors reconstruct training records via targeted queries.
Consequently, Secure Processing Environments must integrate automated gates and specialized technical panels for output verification. This procedure verifies that exported weights, gradients, or tabular summaries lack re-identification vectors before artifacts are permitted to leave the compute boundary.
- Representative functional distribution of operational security components in a secure enclave.
Catalyzing Barcelona's biomedical research ecosystem
For Catalonia, this regulatory clarity offers a decisive strategic advantage. Leading institutions such as the Barcelona Supercomputing Center - Centro Nacional de Supercomputación (BSC-CNS), Hospital Clínic, Hospital Vall d'Hebron, and research clusters coordinated under Biocat now have an explicit operational standard backed by the AEPD to deploy federated clinical research infrastructures.
Barcelona's positioning as a premier European hub for health tech and applied AI is reinforced by its capacity to offer pharmaceutical sponsors and international consortia algorithmic training platforms that strictly comply with the DGA and the EHDS. This clarity accelerates clinical data sharing agreements while resolving longstanding ambiguities in institutional review boards and research ethics committees.
Assessing your organization's technical readiness
Transitioning toward Secure Processing Environments requires a clear audit of analytical systems, spanning data governance, perimeter defense, and confidential computing infrastructure. Knowing where vulnerabilities exist is the prerequisite for designing a compliant AI architecture.
Use this preliminary diagnostic tool to assess how your data pipelines align with the joint standards issued by the AEPD and HDPA.
How do external data scientists or AI developers currently interact with your sensitive datasets?
Strategic outlook for enterprise AI adoption
The release of this joint guidance signals an essential evolution in how European regulators reconcile frontier artificial intelligence development with data sovereignty. By articulating exact technical baselines, regulators have effectively dismantled the false trade-off between privacy compliance and research speed.
Organizations that promptly adapt their data and compute architectures will gain a lasting competitive edge—enabling seamless participation in cross-border clinical initiatives, attracting digital health investment, and unlocking proprietary data capital with unambiguous legal validity.
“The structural transition toward a compute-to-data model establishes an operational standard where datasets never leave their governed perimeter.”



