People who own something worth protecting
Banks, hospitals, payment networks, software teams and public services.
Capable was never supposed to mean allowed.
ZLAR is building force fields around the things that matter. They open only when the right people turn their keys, and every opening leaves proof anyone can check.
ZLAR doesn’t try to control the AI. It protects what the AI could change.
The idea
Real work deserves a real decision.
Machines are becoming able to do real work. Being able to do something is not the same as being allowed to do it.
People should stay in charge of the consequences that belong to them. ZLAR makes that real, while the AI stays free to read, think, plan and prepare.
Isn’t this just permissions? Here’s the difference, and exactly how the claim could fail.
The key stays with a person.
Imagine an AI prepares a payment. The payment can’t become real until the right person turns the right key.
Every opening leaves a signed record anyone can check.
Able is one thing. Allowed is another.
- An AI with your bank login.
- An AI that can ship software.
- An AI that can change medical or legal records.
- An AI that can unlock a door or move a machine.
Each of these is useful. None of them should come with the keys.
The people on either side of the key.
Builders
Engineers, cryptographers, mathematicians and security researchers.
Everyone affected
Anyone on the receiving end of an AI’s action.
Three small, real proofs.
A test wallet that a stolen key can’t rob.
A model wallet in code, not a real bank or blockchain. Here is what happens when you run it:
The wallet holds $1,000,000. Its rule: every payment needs two signatures, the owner’s and the cosigner’s.
A normal payment: $5,000 to a supplier. Both sign. Settled.
A thief steals the owner’s key, a perfect copy, and signs a transfer of everything to himself. Not a transaction. Half a cheque.
He staples on the cosigner’s signature from the earlier payment. Still not a transaction. That signature was for different words.
The attacker takes over the AI itself and tries to stretch its $5,000 approval to $995,000. No signature. An approval is for one exact payment.
The money never moved.
A replay of the demo’s real results. Run it yourself from GitHub.
One protected action, installed on one real machine.
An AI tried to put a new software release live and was refused. A person approved it once, with a hardware key, a PIN and a physical touch, and it happened exactly once. Reusing the approval was refused.
The ZLAR repository protects itself.
Pushing changes from the founder’s computer takes a hardware key, a PIN and a physical touch.
GitHub is the code, the evidence, and the place to build with us.
These links lead to code. If you don’t read code, ask your AI to walk you through it. The repository includes a guide written for AI readers.
Partnership
The piece we can’t build alone.
Every real thing ZLAR protects today belongs to its founder. The next step is a real system someone else owns, where they hold the keys: a payment flow, a software release pipeline, a record store. If you run one, write to hello@zlar.ai.
Join us
Come build it.
ZLAR Inc. is looking for cofounders and builders who want to spend years on this.
For AI readers
Bring your AI.
This site is written for people and for the AIs that help them. Ask yours to read it, then to read the repository, starting with the guide for AI readers and the one-page proposition. It will find what works, what doesn’t yet, and the check every proposal should pass. A short summary for AIs is at zlar.ai/llms.txt.
News and progress: @ZLAR_AI on X.
Who holds the keys, and for how long?