Sequent (also styled "Sequent Research," with the turnstile symbol ⊢) is a nonprofit AI alignment research organization announced on 10 June 2026. It was founded by researchers drawn from the UK AI Security Institute's (AISI) Alignment Team and the startup Timaeus, and is led by Geoffrey Irving, AISI's chief scientist. The organization states that it aims to reach higher a priori confidence in the alignment of advanced AI systems through a portfolio of theoretical and empirical research bets, accelerated by heavy investment in research automation (Source: https://www.alignmentforum.org/posts/AP7YDke5jjY4v3X9Z/sequent-scale-and-automation-for-higher-confidence-in-1).
Overview
In its launch post, Sequent argued that "alignment is not on track" relative to the possibility that artificial superintelligence (ASI) could be developed within a few years, and that the empirical safety programs at AI labs are unlikely to deliver advance confidence that training such systems will go well (Source: https://www.alignmentforum.org/posts/AP7YDke5jjY4v3X9Z/sequent-scale-and-automation-for-higher-confidence-in-1). The organization framed three commitments: aiming at higher confidence through a portfolio of theory and empirics bets, investing in automation to accelerate those bets, and treating theory as a way to unlock more reliable automation. Its stated point of differentiation from AI-lab safety teams is an emphasis on principled reasons—ranging from a better scientific understanding of deep learning to asymptotic safety guarantees under specified training protocols—for believing that alignment observed in controlled settings generalizes to settings that cannot easily be controlled.
Coverage of the launch in the Import AI newsletter on 15 June 2026 described Sequent as a nonprofit launched by the UK AISI alignment team and Timaeus that aims to grow to 40–80 staff and to raise $100–150 million initially (Source: https://importai.substack.com/p/import-ai-461-alignment-is-not-on).
Founding team and structure
The launch post listed an initial founding team:
- Geoffrey Irving — chief scientist at UK AISI; previously at DeepMind, OpenAI, and Google Brain. The post credits him with an early role in establishing reinforcement learning from human feedback (RLHF) at OpenAI and DeepMind and with contributions to scalable oversight via debate.
- Daniel Murfet — head of research at Timaeus, described as having left a tenured mathematics position to direct work applying singular learning theory (SLT) to alignment.
- From the AISI Alignment Team: Alex Holness-Tofts and Jacob Pfau.
- From Timaeus: Jesse Hoogland, Stan van Wingerden, and Marco Cozzi.
(Source: https://www.alignmentforum.org/posts/AP7YDke5jjY4v3X9Z/sequent-scale-and-automation-for-higher-confidence-in-1)
Sequent describes a federated research structure in which a small number of research directors hold substantial autonomy over research direction, culture, and hiring within their areas and report to Irving. It states that its set of research areas is not fixed and will depend on the directors who join. The organization plans a large in-person presence in Berkeley, California, with additional researchers working remotely from London, Melbourne, and elsewhere.
Research agenda
The launch post identified several areas of alignment theory and associated empirical work that Sequent intends to build out or collaborate on, including scalable oversight (amplification, debate, "scientist AI"), learning theory (singular learning theory, computational mechanics), heuristic arguments and low-probability estimation, game theory and agent foundations, and the study of low-dimensional structure in model behavior ("personas"). It described its experience working across AISI Alignment and Timaeus as a motivating example of how distinct research lines can interact.
On automation, Sequent argued that recent improvements in coding agents and in mathematical reasoning by frontier models make automated alignment research more feasible, while cautioning that it is "not confident that automation will work in time." It cited a paper by Bowkis et al., "Automated alignment is harder than you think" (arXiv:2605.06390), as discussing such concerns.
Funding and independence
Sequent stated an initial fundraising goal of $100–150 million, with readiness to raise "at least one order of magnitude more" if it can demonstrate successful parallel exploration of many research directions, and argued that a single large organization is an easier target for funding and better able to amortize the security investments that may be required for frontier-model access (Source: https://www.alignmentforum.org/posts/AP7YDke5jjY4v3X9Z/sequent-scale-and-automation-for-higher-confidence-in-1). It said it had chosen to remain independent rather than join an existing AI lab, citing the value of being able to publicly "yell" if its research surfaces obstacles showing alignment to be difficult, and a concern that safety research inside labs has converged on too few, mostly empirical, bets.
Relationships
- depends-on: UK AI Safety Institute (AI Security Institute) — founding researchers came from AISI's Alignment Team; Irving is AISI chief scientist.
- related: AI Alignment
- related: AI Existential Risk
- related: Superintelligence
- related: RLHF (Reinforcement Learning from Human Feedback)
- related: Timaeus — co-founding group; pioneered singular learning theory for alignment.
- related: Geoffrey Irving — founder and scientific lead.