Good practice

Trilateral KI-Reallabor Pilot — Germany's Simulated AI Regulatory Sandbox for High-Risk Medical AI

Germany · Wiesbaden · See the Germany profile

Germany's Bundesnetzagentur, Hesse's digitalisation ministry and the federal data-protection commissioner simulated an EU AI Act sandbox with two medical-device start-ups, producing a public roadmap ahead of the deadline for member states to stand up real sandboxes.

Start-ups participating in the pilot

~5 months

Pilot duration (May 2025-March 2026)

Details

Maturity: Pilot
Promoter: Bundesnetzagentur (Federal Network Agency), Hessian Ministry for Digitalisation and Innovation, and the Federal Commissioner for Data Protection and Freedom of Information (BfDI)
Period: May 2025–March 2026
Keywords: AI Act implementation, regulatory sandbox, medical device AI, cross-agency governance, high-risk AI compliance

Context

Ahead of the EU AI Act's deadline for member states to stand up regulatory sandboxes, Germany's Bundesnetzagentur, the Hessian Ministry for Digitalisation and Innovation, and the Federal Commissioner for Data Protection (BfDI) ran a trilateral pilot (May 2025-March 2026) simulating a sandbox with two medical-device start-ups developing high-risk AI systems.

Results

The pilot produced a detailed roadmap on how AI Act conformity requirements intersect with medical-device certification, plus lessons on sandbox-candidate viability, participant needs (confidentiality, structured communication, a maturity model), and cross-agency coordination between a federal regulator, a state ministry and a data-protection authority.

Conclusions

The agencies frame the results explicitly as input to designing Germany's forthcoming statutory AI regulatory sandbox rather than a finished, scaled programme; only two companies took part and the output is procedural guidance rather than a quantified outcome.

Implementation

Implementation detail (cost, timeline, staffing, conditions for success) is not yet available for this practice.

Data sources

Where this practice's information was retrieved from, and when.

C-NAPSE web research — accessed 2026-07-13
Bundesnetzagentur (Federal Network Agency), Hessian Ministry for Digitalisation and Innovation, and the Federal Commissioner for Data Protection and Freedom of Information (BfDI) — accessed 2026-07-13
Bundesnetzagentur — Trilaterales Pilotprojekt zur Simulation eines KI-Reallabors erfolgreich abgeschlossen — accessed 2026-07-13
BfDI — Trilaterales Pilotprojekt zur Simulation eines KI-Reallabors erfolgreich abgeschlossen — accessed 2026-07-13
Hessen Digitales — Erfolgreiches Pilotprojekt zur Simulation eines KI-Reallabors — accessed 2026-07-13
Bundesnetzagentur — Begleitforschung zum Pilotprojekt KI-Reallabor (Abschlussbericht, PDF) — accessed 2026-07-13

Quality score: 73/100

How is this calculated?

Evidence: Descriptive / self-reported

Top 16% in this catalogue (246 scored practices)

Scores cluster high, so position within the catalogue is often more telling than the number alone.

Evaluation

Evaluated by C-NAPSE

Score profile — the dimensions behind the score — how it's computed

Evidence of impact / measured public value2/3

The pilot produced a concrete, verifiable deliverable — a detailed compliance roadmap for AI Act/Medical Devices Regulation interaction — documented identically in independent Bundesnetzagentur and BfDI press releases, though it covered only two use cases.

Transparency, fairness & accountability3/3

Findings were published as a joint final report and accompanying PDF (Begleitforschung zum Pilotprojekt KI-Reallabor) by three separate public authorities, each confirming the same facts independently.

Transferability / demonstrated replication2/3

The agencies frame the roadmap explicitly as guidance for the operation of Germany's mandatory statutory AI regulatory sandbox required under the EU AI Act, though it has not yet been applied outside the pilot.

Scalability beyond pilot1/3

Only two start-ups and roughly five months of real-laboratory activity were involved; the pilot is a proof-of-concept for a future full-scale sandbox rather than an operating one.

Governance, capability & sustainability3/3

The project was run jointly by a federal regulator, a state digital ministry and the federal data-protection commissioner, with explicit governance lessons on confidentiality, structured communication and staged maturity assessment.

Cite this

Evidoria. Trilateral KI-Reallabor Pilot — Germany's Simulated AI Regulatory Sandbox for High-Risk Medical AI. Persistent ID: 91232cfb-1edd-4080-9ea6-5674cad29398.

Export as decision brief (PDF)

Similar practices you may find useful

★ 87

DFROG — De Nederlandsche Bank's Machine-Learning-Tested Nowcasting Model for GDP Growth

Netherlands

DNB's DFROG nowcasting model has tracked Dutch GDP growth in real time since 2024, publishing monthly estimates; a decade-long evaluation …

★ 87

Consult — UK Government's AI Tool for Analysing Public Consultation Responses

United Kingdom

Consult, built by the UK's i.AI unit with the Alan Turing Institute, uses generative AI to analyse tens of thousands …

★ 87

vTaiwan — Taiwan's AI-assisted public deliberation platform for policy-making

Taiwan

AI-assisted civic deliberation platform run by Taiwan's PDIS, using open-source Polis AI to map consensus on ~30 policy issues since …

★ 87

Algorithmic Transparency Recording Standard (UK)

United Kingdom

The UK's ATRS is a public register documenting how government algorithmic tools work and affect people. World-first when launched in …