Sequent launches nonprofit to achieve provable AI alignment via theory and automation

Sequent is a new large nonprofit research organization founded by key researchers from the UK AISI’s Alignment Team and Timaeus, including Geoffrey Irving (previously Chief Scientist at UK AISI and ex-DeepMind, OpenAI, Google Brain) and Daniel Murfet (Head of Research at Timaeus, pioneer in applying singular learning theory to alignment). The organization aims to clear a higher bar for AI alignment confidence before the development of artificial superintelligence (ASI). Sequent’s approach is differentiated from reactive lab methods by emphasizing principled theory: a portfolio of bets ranging from better scientific understanding of deep learning to asymptotic guarantees that training protocols converge to safe behavior as parameters are scaled.

Sequent plans to invest heavily in automation to accelerate progress, believing that theoretical insights can provide better filters for promising research directions. The team targets 40-80 full-time equivalents within two years, with a major physical hub in Berkeley and remote contributors from London, Melbourne, and elsewhere. They explicitly state that any single bet could fail, but if even one succeeds, it would provide substantial a priori confidence in aligned outcomes. By focusing on theory-driven automation, Sequent hopes to deliver scalable safety guarantees that go beyond what is achievable through purely empirical or evaluation-based approaches.