Rookih

9 November 2026 · The name is ours again

Intelligence is a lever.
We want it to lift everyone.

Rookih was founded on one conviction: AI can be the greatest lever for good ever built, for humans and for every other being that shares the world with us. Four years after we lost the name, we have it back. The conviction never left.

A small pushSafety, the fulcrumAll beings
01The story

Four years, to the day.

  1. Before 2022

    We built Rookih

    A small team with one belief: AI should be pointed at social good, and “good” has to include more than our own species.

  2. 9 November 2022

    Taken over

    A hostile takeover by one of the large AI companies. It was not the ending we chose, and the terms barred us from using our own name.

  3. 2022 to 2026

    Apart, on mission

    We went to the places where the problem is hardest: risk monitoring, alignment research and international policy.

  4. 9 November 2026

    The name returns

    The restriction expired today. Rookih.com is ours again.

A plain word about the money

Rookih was taken from us in a hostile takeover. It also made us a great deal of money, and we will not pretend otherwise. The money was never the point.

We did not set out to get rich and we did not ask for the sale. The money has been more burden than blessing, and most of it has since gone back out into other work for good. That seemed the only honest use for it.

We started Rookih because we believed AI could be bent toward the good of every being it touches. We believe it more today than we did then, and each of us has spent the years since working on little else.

02The moment

AI is compounding faster than our ability to understand, steer and govern it.

That is the whole situation in one sentence. Whether it goes well is decided on three fronts, and each of us now works on one of them.

I

Readiness

Systems that rewrite and improve themselves, Gödel machines, are moving from thought experiment to engineering goal. We do not think humanity is ready for them yet. Someone has to watch the horizon and say so plainly.

II

Alignment

A powerful system that wants slightly the wrong thing is not a slightly wrong system. Getting AI to reliably want what we mean is unsolved technical work, and it has a deadline.

III

Coordination

No lab and no country can make this go well alone. China, Europe and the United States need shared rules before they need them urgently.

03The concern

The machine that rewrites its own mind.

In 2003 Jürgen Schmidhuber described the Gödel machine: a program that can rewrite any part of its own code, including the part that does the rewriting, once it has proved the change will make it better at its goals. Every improvement makes the next one easier to find.

For two decades this was mathematics on paper. It is now an engineering direction. Research systems that edit their own code and keep whatever scores better have already been published.

Fig. 1An ordinary network passes a signal through once. Wire the output back into the input and it begins to work on itself.

We cannot yet answer basic safety questions about the systems we already have: what they have learned, what they want, how they will behave somewhere new. A machine that redesigns itself turns every one of those open questions into a moving target. Capabilities are compounding far faster than the answers, and that is the wrong order to do this in.

Tim, on why humanity is not yet ready for Gödel machines

Read about Gödel machines

04Ethos

How this goes well.

For all beings

Good that stops at our own species is too small. The animals and ecosystems we share the world with, and any minds we may one day create, belong inside the circle.

Leverage needs a fulcrum

A lever amplifies whatever pushes on it. Safety is not the brake on benefit. It is the point everything else pivots on.

Say what is true

Neither hype nor doom. The risks are real and so is the upside. People can only choose well if they are told both.

Lift widely

If the gains from AI pool with a few, the lever has failed. The measure is how far the benefit reaches, not how high it stacks.

05Team

Three people, three fronts.

I · Readiness

Tim

ThenFounder, led technology
NowTech Lead at an AI existential-risk monitoring company

Whether humanity is ready for self-improving machines. His answer, for now: not yet.

III · Coordination

Paula

ThenHead of people
NowAI policy at a leading tech startup

Coordination between China, Europe and the United States.

II · Alignment

Viktor

ThenHead of mathematics
NowResearch engineer at a leading research lab

Technical alignment, around the clock.

The wider team

Rookih was never only three people. Others built it alongside us, and we remain grateful to every one of them. The three of us were its core team, and we are the ones who carry the name now.

Those who backed us early

Before Rookih could sustain itself, partners including Microsoft, NVIDIA and AWS, among others, wrote our first grants. Their support is the reason there was a company to build at all.

Rookih is still the best thing
we have built together.

We may build together again. Until then the work continues, on three fronts, because we love this cause and believe it can still be won.

Tim, Paula & Viktor