A small agent foundations research group using foundational mathematics to develop rigorous understanding of AI agents and their safety properties.
A small agent foundations research group using foundational mathematics to develop rigorous understanding of AI agents and their safety properties.
People– no linked people
Updated 04/02/26Funding Details
Updated 04/02/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 04/02/26Dovetail is a small, internationally distributed agent foundations research group led by Alex Altair, a former MIRI fellow, MATS scholar, and AI Safety Camp research lead. The group also includes Alfred Harwood (physics PhD from University College London), José Pedro Faustino (mathematics and computer science undergraduate at the University of São Paulo), Dalcy Ku (Harvard undergraduate and 2023 Atlas Fellow), and Daniel C (applied mathematics MSc student at Imperial College London). The group's research is purely mathematical and does not involve machine learning implementation. It draws on dynamical systems, probability theory, information theory, algorithmic information theory, measure theory, and ergodic theory. Core research questions include formalizing the definition of optimization, defining agents and solving the agent structure problem, understanding convergence of abstractions, formalizing value fragility, and assessing utility maximization convergence. Dovetail explicitly positions itself as not doing formal verification or building provably safe AI, but rather developing foundational conceptual clarity in a field that currently lacks consensus definitions. Dovetail publishes its research agenda as a wiki at dovetailresearch.org and releases technical writings on LessWrong and the AI Alignment Forum. In 2024 and 2025, the group ran a research fellowship program offering 10-week to 1-year positions at £36,000 FTE equivalent compensation, funded in part by an ARIA grant. The organization is remote-first globally but encourages UK-based applicants and offers potential in-person meetings in the UK or San Francisco Bay Area. The group is funded by the Long-Term Future Fund (LTFF) and ARIA (the UK's Advanced Research + Invention Agency). There is no public donation page; the organization relies on institutional grants.
Theory of Change
Updated 04/02/26Dovetail believes that many of the core arguments about AI danger can be formalized as precise mathematical theorems, and that doing so would provide clearer safety guidance, improve coordination among researchers, and legitimize policy interventions. Their two key bets are: (1) that conceptual arguments about existential risk from AI can be converted into rigorous mathematical statements, eliminating ambiguity and enabling better coordination; and (2) that safety-relevant properties of AI systems can be understood through general architectural properties — such as optimizer strength and proximity to ideal agent designs — using mathematical tools rather than purely empirical exploration. By establishing foundational paradigms for agent theory and deconfusing researchers about the nature of AI agents, Dovetail aims to enable downstream safety research to be more targeted and effective.
Grants Received– no grants recorded
Updated 04/02/26Projects– no linked projects
Updated 04/02/26Discussion
No comments yet. Be the first to share your thoughts.