New nonprofit Sequent launches to develop alignment techniques for superintelligent AI, raising $100–150M because current lab methods lack principled confidence in safety.

What happened

Researchers from the UK AI Security Institute and alignment startup Timaeus have formed Sequent, a nonprofit research organization aiming to develop alignment techniques—methods to ensure AI systems remain safe and controllable—for superintelligent AI systems. Sequent plans to reach 40–80 fulltime employees within a couple of years and aims to raise $100–150M initially, with preparation to raise at least one order of magnitude more if successful.

Why it matters

Today's AI systems show occasional unpredictable failures, but humans can monitor and fix them. However, as AI systems become smarter and start handling more of their own development through recursive self-improvement, better alignment techniques will be critical to maintain control. Sequent argues that current frontier AI labs use methods that are "essentially reactive" and "do not yield principled insight into if or when they will fail." Independent organizations like Sequent may be better positioned to raise concerns if labs are taking unsafe approaches.

What to watch

Sequent's research will focus on multiple areas—scalable oversight (methods to monitor AI behavior at scale), learning theory, game theory, and personas—with the belief that interactions between these directions could yield breakthroughs. The organization intends to pursue "a portfolio of differentiated alignment bets" rather than a single approach, in contrast to the strategies of major AI labs.