Timaeus is an AI safety research nonprofit whose stated mission is to make "breakthrough scientific progress on AI safety" through applications of learning theory to foundational problems in alignment. Its research applies singular learning theory (SLT) — a body of mathematics drawing on algebraic geometry and statistical physics — to interpretability and alignment, with the aim of understanding how capabilities and values emerge during neural-network training and using that understanding to build alignment interventions (Source: timaeus.co). The organization maintains hubs in Berkeley, Melbourne, and London (Source: timaeus.co).
Founders and approach
Timaeus was founded by Jesse Hoogland, its executive director, and Daniel Murfet, its head of research. Hoogland is described as a cofounder of the developmental-interpretability research agenda and of the broader program of applying singular learning theory to AI safety, which he now pursues with the Timaeus team (Source: timaeus.co). Murfet, a mathematician, left a tenured academic position to direct the application of singular learning theory to alignment (Source: alignmentforum.org).
The group's two primary research programs are developmental interpretability — the study of how internal structure forms over the course of training — and "Spectroscopy," an interpretability methodology that uses susceptibility-based methods grounded in weight space, in the tradition of statistical physics and Bayesian statistics, rather than in activations as sparse autoencoders and probing techniques do (Source: timaeus.co).
Activities
Beyond its core research, Timaeus runs field-building activities aimed at drawing mathematicians and physicists into AI safety. These have included the ODYSSEY conference for theoretical AI-alignment researchers (Berkeley, August 2025), a "Mathematical Science of AI Safety" focus period at the University of Sydney with Monash University and the Sydney Mathematical Research Institute (November–December 2025), and a Research Fellows Program launched in April 2026 for tenured faculty and senior researchers who collaborate with Timaeus while keeping their academic positions (Source: timaeus.co; timaeus.co).
In June 2026 several Timaeus researchers, including Hoogland, helped found the alignment nonprofit Sequent, which combines AISI's Alignment Team with the Timaeus singular-learning-theory group; Murfet was listed as head of research at Timaeus in Sequent's launch materials (Source: alignmentforum.org).
Relationships
- related: Sequent (Sequent Research) — co-founding group; supplied the singular-learning-theory research line.
- related: AI Alignment
- related: Mechanistic Interpretability