Sequent: scale and automation for higher confidence in alignment

2026-06-10 · Source: AI Alignment Forum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, long

Summary

Sequent, a new large nonprofit research organization, launched on June 10, 2026, with the goal of achieving higher confidence in AI alignment before artificial superintelligence (ASI) is developed. Founded by researchers from the UK AISI's Alignment Team and Timaeus, including Geoffrey Irving and Daniel Murfet, Sequent aims for 40-80 full-time equivalents within two years. It differentiates itself from AI labs by emphasizing theoretical proofs and automation to ensure alignment generalizes beyond controlled environments. Sequent plans to invest heavily in automated research, using frontier models for informal mathematics and experiments, while building error-correction mechanisms. The organization will adopt a federated structure to preserve research diversity and aims to raise \$100-150M initially, with plans for significantly more funding to support large-scale automation and amortize security costs. Sequent maintains independence to pursue a diverse portfolio of theoretical and empirical bets, including scalable oversight, learning theory, and game theory, and to voice concerns if alignment proves difficult.

Key takeaway

For AI Scientists and Research Scientists focused on alignment, Sequent's launch signals a critical shift towards theory-driven automation. You should consider how integrating theoretical insights with empirical methods can enhance confidence in your alignment protocols' generalization. Explore opportunities to collaborate with or join organizations like Sequent that prioritize diverse research portfolios and automated experimentation to accelerate robust alignment solutions.

Key insights

Sequent aims to achieve higher AI alignment confidence through a portfolio of theory-driven, automated research bets.

Principles

• Theory enables higher automation.
• Scientific reputation attracts talent.
• Federated structure preserves diversity.

Method

Sequent will use frontier models for informal mathematics and experiments, integrating error-correction and human research taste to accelerate alignment progress.

In practice

• Integrate learning theory with scalable oversight.
• Apply game theory to reachable equilibria.
• Use personas for model behavior insights.

Topics

• AI Alignment
• Automated Research
• Singular Learning Theory
• Scalable Oversight
• Deep Learning Theory
• Game Theory

Best for: Investor, AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• Apr 14, 2026AlignmentAutomated Alignment Researchers: Using large language models to scale scalable oversight — AI models can autonomously accelerate alignment research by significantly outperforming human baselines in specific, measurable tasks.
• Quantifying Faithful Confidence Expression in Large Reasoning Models — Faithful confidence expression is a distinct, significant challenge for Large Reasoning Models, requiring new quantification methods.
• Advice for making robust-to-training model organisms — Fragile AI misalignment model organisms hinder technique evaluation; robustness requires specific training and design choices.
• Alignment as Institutional Design: From Behavioral Correction to Transaction Structure in Intelligent Systems — AI alignment should shift from behavioral correction to institutional design, fostering aligned behavior through internal economic structures.
• The inaugural Redwood Research podcast — AI safety requires pragmatic strategies, acknowledging both technical challenges and organizational realities, to manage escalating risks.
• AlignmentMay 8, 2026Teaching Claude whyNew research on how we've reduced agentic misalignment. — Teaching AI models ethical principles and reasoning generalizes alignment better than mere behavioral demonstrations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Alignment Forum.