Loek Verdonk

I am an independent AI evaluation researcher in the Netherlands and the project lead behind No Silent Landing. I turn broad claims about AI systems into concrete, reproducible tests. My work focuses on evidence lineage: whether the source reviewed, the evidence recorded, the human approval given, and the object ultimately used still correspond. I build adversarial test cases, negative fixtures, conformance tests and clear PASS / FAIL / UNKNOWN boundaries for AI systems and AI-assisted workflows. Typical questions include: Is an answer actually grounded in the cited source? Is the object a person approved the same object later activated? Does a reported safety block really prevent downstream use? My background is in operations, entrepreneurship, production and team leadership rather than a traditional academic or ML-engineering path. I entered AI safety through hands-on evaluation work and practical quality control. My research is deeply AI-assisted. I remain responsible for the observations, evidence, scope, claims and final human decisions. I am especially interested in AI evaluation, human oversight, reproducibility, evidence integrity and inspectable AI governance.

Personal Website

https://profile.indeed.com/p/loekv-mpq84xn

Projects

No grants recorded.