Superalignment

Superalignment

We’re building the first practical solution to superalignment.

Soon we will build machines smarter than the people meant to supervise them. Superalignment is the problem of checking work you cannot out-think. We answer it with evidence instead of trust: nothing acts until it has earned the right to.

A periodic three-body orbit What am I looking at?

The product

Automate your organization with superintelligence. On your terms.

Praxis is what our research looks like once it has to survive a real operation.

The capability to automate most of an operation already exists. What is missing is a way to hand the work over without handing over the judgment, so the person accountable for it can actually say yes. Praxis is built for that gap.

Four parts. One place that has to answer for all of them.

The research

Convergence Programming

Iterative, World-Grounded Alignment of Lossy Human Intent with Executable Behavior.

The paper names a failure we can now measure: false convergence, where visible signals say a system is ready while hidden requirements remain unsatisfied, unmeasured or untested. It argues that readiness is information-limited, meaning a system cannot certify distinctions its trajectory has not extracted from the world, and it sets out what would have to be true to certify one anyway.

The full paper is being prepared for release. Ask us for a copy in the meantime.