Aidan O'Gara
This stream will focus on compute governance, including data center inspections, export controls, and distributed training.
Stream overview
I'm interested in supervising projects on compute governance. For example, here's a recent project proposal I wrote about data center inspections:
"How can nations verify compliance with international agreements on frontier AI development? This project will explore one potential verification mechanism: inspections of data centers.
First, we'll analyze the concrete affordances that inspections could offer: physical inspection of chips, interviews of data center staff, camera monitoring, tapping of communication lines, and other information gathering mechanisms. Then, we'll look at historical case studies of inspections to verify other international agreements, such as agreements on nuclear, chemical, and biological weapons, to draw lessons that can be applied to AI. Finally, we'll discuss the various challenges faced by data center inspections, and provide concrete steps that policymakers and researchers can follow to improve the prospects for their success."
Other projects I'd be interested in supervising could include a holistic assessment of US export control policy on AI, or a literature review on distributed training and its implications for compute governance. I see these projects as contributing to the ability of governments to understand and regulate the development of advanced AI, thus limiting its downsides.
Mentors
Aidan O'Gara is a DPhil candidate in Oxford's Department of Engineering Science and a DPhil affiliate of the Oxford Martin AI Governance Initiative. In 2025, he co-authored a research agenda on hardware-enabled mechanisms for AI governance.
Mentorship style
Fellows we are looking for
Project selection
The Winter 2026 cohort offers a wide range of research streams led by experts across AI alignment, interpretability, governance, and safety. Each stream provides its own research agenda, methodology, and mentorship focus.
