Derck Prinzhorn
AI Safety & Security
About
I'm a Member of Technical Staff at Exponential Security Labs, where I work on AI safety and security for agentic systems.
My research focuses on AI control, automated attacks, monitoring, and oversight of increasingly capable models. I recently completed my MSc in Artificial Intelligence at the University of Amsterdam, with thesis research at the Max Planck Institute for Intelligent Systems and ELLIS Institute Tübingen.
I also run Prinzhorn Solutions, helping organizations manage the risks of adopting AI systems. Previously, I worked as a Research Engineer at Aithos, as AI Architect at the Dutch Police, and co-founded Wisr. I graduated cum laude from my bachelor's, receiving the Amsterdam AI Thesis Award for my work on uncertainty quantification.
Experience
Exponential Security Labs
Aug 2026 – present
Member of Technical Staff
Building self-improving red-teaming and guardrail agents to secure agentic systems.
Max Planck Institute
Jan 2026 – Aug 2026
Research Intern
Research on AI control under supervision of Maksym Andriushchenko.
Prinzhorn Solutions
Apr 2025 – present
Founder
Helping companies understand and manage risks associated with adopting AI systems.
Aithos
Apr 2025 – Jan 2026
Research Engineer
Worked on evals for AI value systems and moral competence.
Wisr
Sep 2024 – Oct 2025
Co-Founder
Worked on a startup helping teachers save time with AI grading.
University of Amsterdam
Jan 2024 – Feb 2025
Research Intern
Conformal prediction for time series; 3D diffusion models for radiotherapy dose prediction; physics benchmarking in video generation models.
Dutch Police
Apr 2023 – Apr 2025
AI Architect
Defined reference architectures for AI, MLOps, and AI security.
Publications
Highlights / News
2026
Aug
Joined Exponential Security Labs as Member of Technical Staff.
Jul
Released ResearchArena, a framework for evaluating sabotage and monitoring in automated AI R&D.
Mar
Participated in the Apart Research AI Control Hackathon, producing Controlling the Researcher.
Jan
Started thesis research on AI control at MPI-IS and ELLIS Institute Tübingen.
2025
Jul
HIVE paper accepted at the Beyond Euclidean Workshop, ICCV 2025.
Apr
Joined Aithos for pluralistic alignment research. Founded Prinzhorn Solutions.
2024
Dec
Presented FairAC reproduction as a poster at NeurIPS 2024.
Sep
Co-founded Wisr (EdTech). Presented conformal time series paper at COPA in Milan.
Jul
Joined SPAR, working on AI control with Aryan Bhatt from Redwood Research.
May
FairAC reproducibility paper accepted at TMLR. Conformal decomposition paper accepted at COPA.
2023
Nov
Received the AmsterdamAI thesis award for uncertainty quantification in time series.
Sep
Started MSc in Artificial Intelligence at the University of Amsterdam.
Projects
Mar 2026
Controlling the Researcher: AI Control Evaluations for Automated AI R&D
Control evaluations for AI agents doing ML research. Subtle sabotage embedded in artifacts evades nearly all monitors, while obvious side tasks are reliably caught.
Oct 2025
HIVE: Hyperbolic Visualization Explorer
Interactive dashboard for exploring hyperbolic embeddings with curvature-aware projections and multiple interaction modes. Published at ICCV 2025.
Jul 2025
Injecting Image Guidance into Diffusion Models
Guide Stable Diffusion with both text and a reference image at inference time, without retraining. A lightweight aligner bridges the image-text embedding gap.
Sep 2024
Transformer-Based Radiotherapy Dose Prediction
Deep learning for radiotherapy dose prediction in head and neck cancer. Extended UNETR with physics-informed losses and a sequential RNN decoder (~10% improvement).