Bio
I’m a Research Fellow with the Safe AI Forum, where I work on concrete proposals for international agreements.
I previously built an R&D agenda for societal resilience with the UK’s Advanced Research and Invention Agency, built a loss of control demo for the UK government with GovAI, and researched open-source policy at ERA. I have an MPhil in AI & Ethics from the Center for the Future of Intelligence, Cambridge.
Projects
Research Direction: Realistic US-China Agreements on AI Risk
This workstream approaches international AI governance from a pessimistic, political-realist perspective.
Specifically, strategic pictures like Plan A by AIFP or work by MIRI often portray a future where the increasing saliency of AI risk and sympathetic political will create the circumstances for effective international action to mitigate AI risk. In contrast, this workstream assumes that 1) the diplomatic relationship between Washington and Beijing remains volatile, 2) states remain uncertain, conflicted, or unconvinced by key threat models around AI risk, even as models demonstrate (jaggedly) superhuman capabilities and warning shots emerge, and 3) states generally have limited staffing capacity for international diplomacy and adapting to rapid AI capabilities progress.
The goal for this workstream is to identify, scope and roadmap concrete proposals for mitigating AI risks that the US and China might agree to under this worldview.
Projects we might work on include:
- Fine-graining clauses for agreements. You might work with me to specify an exact phrasing or clause structure for an agreement to restrict undesirable practices in AI R&D.
- Scoping new agreement ideas. You might explore whether the US and China should engage in information sharing about control protocols, or build up their security standards to prevent non-state actors from obtaining model weights.
- Building better models of the strategic context. You might seek to shape SAIF’s strategic direction by conducting deep-dives into how rare earth metals, UAVs, different AI takeoff scenarios or nuclear deterrence will impinge on key issues regarding AI diplomacy.
We might also work on something different, and the scope of this project could change to match your ambition and experience.
The success criteria for work from this project will be heavily biased towards delivering concrete, actionable insights that can direct international AI governance work in 2027.
What I'm looking for in a Mentee
You like to learn fast and reason under uncertainty.
Bonus points if you have: a Substack or public writing on international (AI) governance, a technical background related to AI verification or security, experience working in diplomacy or foreign affairs, or speak Mandarin.
As a mentor, I have three main goals: to help you do work that will advance the field, to improve your ability to identify promising research directions, and to support you in moving towards a productive career in the field after the fellowship should you choose to do that.
What I'm Like as a Mentor
I tend to vary my management styles depending on the personality, seniority, and goals of mentees. That said, I will generally try to maximise the value of our weekly meetings, set clear expectations for deliverables, and have a low bar for communication.