Manj Chenna, who builds trust infrastructure for AI in Amsterdam Manj ChennaAmsterdam

I build trust infrastructure for AI.

The values a system runs on, the human it can reach before it decides alone, and independent evidence of what it actually did. You need all three.

This page is alive. Behind it is Conway's Game of Life, reseeded on every visit, so what you are looking at has never existed before. Move your cursor and you seed more. Play it.

ValuesSanctityThe human values AI should follow.live↗JudgmentHumanChainA named human it can reach, before it decides alone.early access↗ EvidenceModelometerHow models actually behave, and when they change.live beta↗
Scroll
Manj Chenna

How I decide

I will not let a system decide until it answers three questions.

They are separate questions and the answers come from three different places, which is why I ended up building three things instead of one. They are separate products at different stages, not one finished system.

  • Whose values is it running on?

    Somebody already chose what the system should want. I want that choice written down, argued over and open to challenge, rather than left to whoever shipped last. At Sanctity we write the candidate lines and public answers help decide which are proposed: one public constitution today, with individual and group versions intended but not yet available.

    Sanctitylive ↗

  • Who decides when it should not?

    Every capable system reaches a question it has no business answering alone. The useful move is to stop and ask a named human, and to make that pause cheap enough that people actually use it. HumanChain is that pause, and it is in early access.

    HumanChainearly access ↗

  • What actually happened?

    Claims about a model age badly. Modelometer checks, on a schedule, what was really serving and what changed underneath it, and seals the result before anyone is allowed to read it. It is in live beta: early runs have caught endpoint swaps, and we are still validating how reliably.

    Modelometerlive beta ↗

Take one away and the other two stop meaning anything. That is the whole thesis, and everything on this page is a consequence of it.

Experiments

I test ideas by making them playable.

An argument you can lose to is worth more than an argument you can only read. Two of these are toys. The third one is not, and it will probably annoy you.

Game of Life↗Conway's four rules, unchanged since 1970. Simple arithmetic that behaves like it is thinking, which is why it runs behind everything I build.Ecosystem Tank↗Agents, scarcity and incentives in a closed tank. Watch cooperation appear on its own, then watch something learn to exploit it.Rubber Stamp↗You are the human in the loop. Twelve real decisions an AI has already made, seconds each, and four of them are wrong. You get a score.

Writing

I work the argument out in public.

No ghostwriter and no content calendar. Every claim ships with what would change my mind, and if a position of mine is wrong, this is where it gets caught.

Justified epistemic verification, or how one model stopped being enoughRead →Human oversight is mostly theaterRead →The Game of Life, explainedRead →Why my logo is a gliderRead →
All essays →

Who I am

Where all of this comes from.

Manj Chenna, founder of Sanctity, who builds trust infrastructure for AI in Amsterdam

I am Manj Chenna, and I build trust infrastructure for AI: the values a system runs on, the human judgment it can reach before it decides alone, and independent evidence of what actually happened. I do that from Amsterdam, inside the EU AI Act, while most of the frontier conversation happens nine thousand kilometres away in San Francisco. That gap is my vantage point. I am not theorising about AI governance, I am shipping under it.

It did not start with an engineering degree. It started at nine, in a school computer lab in 1990s India, years before any of this had a name. The living grid behind this page and the glider in my logo are the same idea I have been circling ever since: a few simple rules, chosen by a few people, quietly growing the world the rest of us live in. The longer story.

I am an engineer who spent a career being hard to impress, and this is the first technology that genuinely awed me. I decided awe deserved responsibility, not applause. Good people recuse themselves from power over things like this, and that is how it ends up raised by the other kind, so I stopped recusing myself.

Sanctity is a heavy word to name a company after, and I mean it as a constraint rather than a vibe. There are lines we do not cross even when crossing them is faster or more profitable. We do not ship a decision a person cannot refuse, and we will not strip out the human override switch, not for any contract. A product is defined as much by its no as its yes.

The second source is geography. From that lab I went on to live in ten countries and travel to all seven continents, which is exactly why I distrust any single room deciding the values a model carries for everyone else.

⌂ Home Lived Travelled Drag to spin

Find me

If you build, regulate or write about who controls AI.

LinkedIn X Email Sanctity New essays by RSS, no list, no funnel