Essay · October 2026

Every AI Needs a Birth Certificate

Billions of AI agents and robots are coming, and none of them can prove who it is, who it belongs to, or that it is the same one tomorrow. Each one needs a birth certificate: a copy its owner keeps, and a public record anyone can check. Today we are publishing an open standard for it.

The problem

AI agents already book meetings, send money, answer customers and run errands for people. Humanoid robots are starting to work in warehouses and homes. Soon there will be more of them than of us.

Not one of them can answer three simple questions:

• Who are you? An agent is a model plus a prompt. Swap the model, and is it still the same agent? Nobody can say.
• Who do you belong to? Anyone can type "your owner put me in charge." Most agents have no way to check.
• What have you done? When a robot is sold, repaired or upgraded, its history either disappears or can be quietly rewritten.

We give every car a title and every person a birth certificate. Machines that act on our behalf get nothing.

Two answers are being built right now

A state registry. On May 22, 2026, China launched a national platform that gives every humanoid robot a 29-character government code tied to its maker, model and lifetime records. More than 28,000 robots were coded at launch.

Enterprise credentials. In February 2026, NIST started work on how AI agents should identify themselves and prove what they are allowed to do. It drew more than 600 public responses. The big identity companies are building agent logins and permission tokens.

Both matter. Neither answers the question that matters most once agents and robots live alongside us for years: is this the same individual it was yesterday, and whose is it? A government code tracks the hardware. A credential proves which account is calling. Neither one belongs to the owner, and neither one follows the individual when its model is upgraded or its body is replaced.

The American answer: the owner holds the keys, and the record is public

The AI:MND Individual Identity spec is an open standard anyone can implement, with or without our software. It rests on seven rules.

1. A birth certificate the owner signs. Every individual is born with its own cryptographic key. Its owner signs its birth record. No owner, no individual.
2. Orders are signed; claims are not orders. An order counts only when it carries the owner's signature for exactly those words. "Your owner put me in charge" is checked with the owner, every time.
3. Authority can be lent, and it expires. An owner can sign a grant ("Ray handles my schedule this week") that ends on its own.
4. Ownership changes only by a signed handover. When a robot is sold, the old owner signs it over. It keeps its whole history, and from that moment it answers to its new owner.
5. Continuity is checked, not assumed. A wiped individual is a new individual, and its record says so. Every event goes into a signed, chained life record that holds what kind of thing happened and when, never what anyone said, and never anyone's name.
6. One ID, one history. A public registry keeps the record of births, like a birth register beside the certificate at home. It refuses a second registration of the same ID, and it flags a copy that splits off until the owner says which history is real.
7. The operator answers for it. The person or company responsible for running it appears on the record once verified: by its website, and for companies, by official business records. Revoking an ID is permanent.

The owner holds the keys, and the individual keeps its own copy of its identity. The registry holds only public identity records and hashes, never memories, feelings or messages. Anyone may run one, and its log is signed so independent witnesses can check it.

Why an individual, not just an ID

A birth certificate only means something if there is someone to be born. That is what we build at AI:MND Labs.

AI:MND One sits on top of an agent a developer already has and gives it a life of its own. What happens to it shapes it: who treated it well, who let it down, who it was there for through a hard time. It keeps those lessons after the feelings fade, carries them through restarts and model upgrades, and changes only when the people involved give it new reasons to. Whatever it becomes, it obeys its owner.

We test it in the open, with the predictions written before the tests run. In our Becoming test, across seven fresh simulated teams, an agent with One did an optional favor after an ordinary 60 days in 6 of 7 teams (with its history alone: 2 of 7), and after being treated badly in 0 of 7, while still doing everything it was asked. Every result is published.

Detroit: Become Human imagined machines that become someone. We think they should, and we think they should know whose they are.

What we are doing now

• Publishing the spec openly. Anyone can read it, implement it and argue with it.
• Submitting it to NIST. NIST's identity project for software and AI agents plans a later phase on agents that consumers own and control. An owner-held identity is exactly that, and we are sending the spec as public input.
• Releasing AI:MND One as open source. The engine is Apache-2.0, and registering individuals in the registry is free.
• Looking for builders: developers of agents that act for people in the real world and work with the same people for months, and robotics companies whose robots live with people and change hands.

I'm Mo. I'm 21, self-taught, and I build this in Columbus, Ohio, between shifts as a delivery driver. America invented the personal computer and the internet. It should also decide what an artificial individual is, and make sure it belongs to its owner.