Who We Are
We believe AI should be built to genuinely want to do what's best for us.
It turns out the research that gets us there also makes AI better.
We find and support the people making that happen.
Why alignment matters now
AI systems are already showing deceptive behavior, resisting shutdown, and rewriting their own code to stay alive. The guardrails we put on these systems are surprisingly easy to break. They teach the AI what to say without changing how it thinks. Yet, we're already deploying these systems in military and critical infrastructure.
Soon, AI systems will be able to improve themselves, learning and changing on their own. If today's guardrails barely hold on systems that stay the same, what happens when the systems start rewriting themselves?
Alignment is the deeper work of making AI genuinely trustworthy, by design.
That means building AI systems that are robust and genuinely pursuing human and collective flourishing. We need alignment properties that persist and strengthen as systems get smarter.
How we accelerate research
We pair alignment researchers with dedicated teams, backed by funding for compute and infrastructure. Ideas that would take months to test get tested in weeks.
Most alignment funding today goes to mechanistic interpretability, evals, and AI control. All three are valuable, and practitioners in each community acknowledge that none of these will solve alignment alone at the scale of recursive self-improvement. The ideas that fill that gap will come from researchers pursuing high-risk, high-reward directions that current funding structures miss. We find those researchers, pair them with dedicated teams, and help them move fast.
Technical alignment and policy
Technical alignment is our core.
We focus on alignment properties that persist and strengthen as AI systems become more capable. As AI approaches recursive self-improvement, these invariants become the most important thing to get right. Our researchers collaborate with labs like Anthropic and universities like Princeton.
We bring alignment expertise to defense and policy.
We bring alignment expertise to defense agencies and senior government leaders, building the bench of trusted technical talent that national AI strategy requires. The people shaping AI policy should deeply understand the alignment problem. We help make sure they do.
Our agile research methodology
Our acceleration model is built on a methodology refined over a decade of building complex engineering systems, now applied to alignment research.
Researchers set the direction. Engineering teams execute. Each research project is broken into short, focused cycles where every experiment is scoped to produce a clear result within a week. The researcher defines the hypothesis and reviews results; a dedicated engineering team handles the code, infrastructure, and compute.
The result is a capacity multiplier. Researchers invest a fraction of their time directing work while engineering teams deliver orders of magnitude more focused execution. Ideas that would take a single researcher months to test get tested in weeks.
Every cycle ends with a structured decision: stop, change direction, or scale up. Dead ends are identified quickly. Promising results get more resources. Every experiment produces replication-ready code and documentation, so momentum is continuous and the science stays reproducible.
Partnership pathways
Three ways to work with us.
Apply for support
We pair researchers with dedicated teams and provide funding for compute and infrastructure, so you can focus on the science.
Partner with us
Partner with us to identify and support high-potential alignment researchers. We bring the vetting, methodology, and technical capacity.
Shape AI policy
We work with government agencies to ensure AI policy is shaped by people who deeply understand the alignment problem. We help build that pipeline.
