Research agenda
Last updated: October 2026. If something here interests you — as a fellow, student, research assistant, or collaborator — please get in touch: sa478@cam.ac.uk.
Papers in progress
Strategically stable pareto-topia
What arrangements for the distribution of power and benefit in a world with advanced AI are both strategically stable (no major actor has an incentive to defect) and broadly acceptable (most actors prefer them to continued racing)? Existing “good outcome” visions tend to assume away the stability problem; existing stability analyses tend to assume narrow actor sets. Status: concept stage. A collaborator could own: mapping the existing proposal space against explicit stability criteria; game-theoretic or simulation probing of candidate arrangements.
Reflections on funding systemic AI safety
What did designing and running a national Systemic Safety grants programme teach us about how to fund safety-relevant R&D, and how should governments and philanthropists fund it differently? Connects to my pre-AI research on science-funding mechanisms. Status: skeleton exists. A collaborator could own: a structured retrospective of the grants portfolio; comparison with other safety-funding mechanisms; interviews with grantees and programme staff.
AI agents (written at AISI, unpublished)
A full draft on AI agents written during my time at the UK AI Security Institute, currently seeking a publication route. A collaborator could own: updating the draft against the 2026 agents landscape; shepherding publication.
Papers in planning
Safety interventions beyond the model
Most safety work targets the model; most deployed risk runs through everything around it. A research line on harness- and tool-level interventions, multi-agent and agent-human interfaces, and organisational and sector-level safeguards. First output: a taxonomy and research agenda paper.
Balance of power in a post-AGI world
What does a multipolar, broadly acceptable post-AGI order look like, and what near-term moves make it more likely? Candidate outputs: a scenario-mapping paper; a simulation-gaming study using TSR-style exercises as the probe.
Routes to middle-power relevance
How should middle powers stay relevant through the AI transition? The two levers I find most tractable: position in the compute supply chain, and AI-for-science capacity. Most existing work is US-centric or generic; this line aims at rigorous, country-specific analysis.
Wargaming and roleplay for better AI decision-making
TSR has run 90+ workshops with 700+ participants since 2022; the methods evidence base lags far behind the practice. Planned: a methods paper distilling what simulation gaming can and cannot tell us about AI futures.
What I’m looking for
- Fellows and students (GovAI Winter Fellowship, ERA, and similar): own a well-scoped subset of one of the papers above, with weekly supervision.
- A research assistant (via CSER/ITH): literature mapping, interview logistics, draft wrangling across the portfolio.
- Commissioned research teams: contained, well-specified chunks — for example, the middle-power compute-supply-chain mapping.
- Co-authors with complementary expertise: game theory, science-funding policy, standards and audit practice.