Research

My goal is to help build safe, secure AGI/ASI through (hardware) verification, benchmarking/evaluations, and governance that bridges technical rigour with real-world impact. Before that, I built everything from systems used by millions to high-throughput trading platforms.

I’ve been fortunate to work with (and be supervised by) Tobin South & Ben Bucknall (Pivotal Research), Prof. Alastair R. Beresford & Daniel Hugenroth (Cambridge), and Prof. Nic Lane & Bill Marino (Cambridge). During undergrad, I worked under Prof. Georg Carle on security and anonymity of mix networks, and co-authored a climate-AI chapter with Prof. Isabell Welpe.

Selected Work

• AIReg-Bench: Benchmarking Language Models That Assess AI Regulation Compliance (link) B. Marino, R. Hunter, Z. Jamali, M. E. Kalpakos, M. Kashyap, J. Hinton, C. Schnabl, et al. ICLR 2026 (Under Review)
• Attestable Audits: Verifiable AI Safety Benchmarks Using Trusted Execution Environments (link) C. Schnabl, D. Hugenroth, B. Marino, A. R. Beresford ICML TAIG Workshop ’25, in commercial use (under NDA) where I serve as an advisor.
• Cannot or Should Not? Automatic Analysis of Refusal Composition in IFT/RLHF Datasets and Refusal Behavior of Black-Box LLMs (link) A. von Recum, C. Schnabl, G. Hollbeck, S. Alberti, P. Blinde, M. von Hagen NeurIPS SFLLM Workshop ’24, Dataset has informed the NVIDIA garak LLM vulnerability scanner.
• Private Group Management for Mix Networks (link)
  C. Schnabl Published and presented at TBC 2024