K01

K01 ehf. is a health AI company in Reykjavík, Iceland (EU/EEA), founded in 2023 by Dr Arinbjörn Kolbeinsson and Dr Benedikt Kolbeinsson, both with PhDs from Imperial College London. K01 builds generative models of how patients’ health develops over time, and uses them to simulate patients, test clinical AI and share health data safely.

What K01 builds

• Patient simulation: multi-modal patient trajectories (electronic health records, 12-lead ECG, medical imaging, physiological and wearable signals), including counterfactual trajectories under different treatments or interventions.

• Simulated clinical environments: scored scenarios and a simulated hospital (FHIR, agent tool interfaces such as MCP) for testing AI agents, clinical decision support and LLM-based tools before and after deployment. K01 Trajectory, a reinforcement-learning environment for medical-reasoning agents, is in development.

• Synthetic patient cohorts: K01 Cohort generates realistic cohorts on demand through an API and an MCP server, with outputs in FHIR R4 and R5. Models are trained under differential privacy, so no individual can be recovered from the output.

• Models trained on partner data: K01 Lab trains a bespoke generative model inside a hospital’s or registry’s own environment under differential privacy. The patient data stays with the partner and only the trained model leaves.

• Privacy testing of data releases: an offline AI red team that attacks synthetic, anonymised or pseudonymised datasets (membership inference, attribute inference, linkage and re-identification) and reports the risk, in support of GDPR anonymisation assessments and secondary use under the European Health Data Space.

What K01 can contribute to a consortium

• Synthetic and privacy-preserving data to unlock sensitive clinical data for development, testing and sharing across sites, including under the EHDS

• Patient simulation and counterfactual trajectories for in-silico studies and what-if analysis of treatments and care pathways

• Evaluation and validation of clinical AI: performance, robustness, bias, reproducibility and trustworthiness metrics, in line with the EU AI Act and the Ethics Guidelines for Trustworthy AI

• Testing environments for clinical decision support and AI agents, with scenario libraries and benchmarks

• Privacy risk assessment of shared datasets and models

• Generative foundation models for physiology and time series (ECG, vital signs, wearables)

• Software engineering for research outputs: APIs, FHIR interoperability, open-source tools

K01 typically leads or co-leads a work package or task on synthetic data, simulation, AI evaluation or privacy, alongside clinical, academic and industry partners.

Track record

• Peer-reviewed workshop papers at ICLR, ICML and NeurIPS 2026, on differential privacy for synthetic data, conditional ECG generation, synthetic training data, time-series foundation model benchmarks and AI agents in the life sciences

• Prior work by Arinbjörn Kolbeinsson includes Terminal-Bench (ICLR 2026), a widely used benchmark for evaluating AI agents, and Homekit2020 (CHIL 2023), a benchmark for health prediction from wearable data

• Research grants from the Icelandic Technology Development Fund (Rannís) since 2024, and a EuroHPC AI Factory compute allocation on LUMI

• Collaborations with a primary-care provider, a university hospital haematology group and a European rare-disease diagnostics consortium

• Industry research experience at Samsung AI and Evidation Health (Arinbjörn Kolbeinsson)

Topics of interest (Horizon Europe Cluster 1, 2027)

• TOOL-04: Virtual Human Twins for integrated clinical decision support

• TOOL-01: AI predictive biomarkers for chronic non-communicable diseases

• CARE-02: Reducing adverse drug reactions from multiple medications

• TOOL-08: Next generation of frontier AI models for healthcare

• DISEASE-01: Innovative healthcare interventions for non-communicable diseases

• IND-02: Portable point-of-care diagnostics

• DISEASE-14: Clinical trials for neurodegenerative diseases

• ENVHLTH-02: Climate-related exposures in the human exposome

Keywords: synthetic data, synthetic patients, patient simulation, patient trajectories, counterfactuals, in-silico, simulation environments, AI agents, clinical decision support, AI evaluation, benchmarking, trustworthy AI, bias, robustness, differential privacy, privacy-preserving AI, anonymisation, re-identification risk, GDPR, EHDS, secondary use, AI Act, generative AI, foundation models, diffusion models, time series, ECG, wearables, medical imaging, electronic health records, FHIR, primary care, haematology, rare diseases.